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 #include <asm/barrier.h>
21 
22 #include "kvm_util.h"
23 #include "test_util.h"
24 #include "guest_modes.h"
25 #include "processor.h"
26 
27 #define DIRTY_MEM_BITS 30 /* 1G */
28 #define PAGE_SHIFT_4K  12
29 
30 /* The memory slot index to track dirty pages */
31 #define TEST_MEM_SLOT_INDEX		1
32 
33 /* Default guest test virtual memory offset */
34 #define DEFAULT_GUEST_TEST_MEM		0xc0000000
35 
36 /* How many pages to dirty for each guest loop */
37 #define TEST_PAGES_PER_LOOP		1024
38 
39 /* How many host loops to run (one KVM_GET_DIRTY_LOG for each loop) */
40 #define TEST_HOST_LOOP_N		32UL
41 
42 /* Interval for each host loop (ms) */
43 #define TEST_HOST_LOOP_INTERVAL		10UL
44 
45 /* Dirty bitmaps are always little endian, so we need to swap on big endian */
46 #if defined(__s390x__)
47 # define BITOP_LE_SWIZZLE	((BITS_PER_LONG-1) & ~0x7)
48 # define test_bit_le(nr, addr) \
49 	test_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
50 # define __set_bit_le(nr, addr) \
51 	__set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
52 # define __clear_bit_le(nr, addr) \
53 	__clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
54 # define __test_and_set_bit_le(nr, addr) \
55 	__test_and_set_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
56 # define __test_and_clear_bit_le(nr, addr) \
57 	__test_and_clear_bit((nr) ^ BITOP_LE_SWIZZLE, addr)
58 #else
59 # define test_bit_le			test_bit
60 # define __set_bit_le			__set_bit
61 # define __clear_bit_le			__clear_bit
62 # define __test_and_set_bit_le		__test_and_set_bit
63 # define __test_and_clear_bit_le	__test_and_clear_bit
64 #endif
65 
66 #define TEST_DIRTY_RING_COUNT		65536
67 
68 #define SIG_IPI SIGUSR1
69 
70 /*
71  * Guest/Host shared variables. Ensure addr_gva2hva() and/or
72  * sync_global_to/from_guest() are used when accessing from
73  * the host. READ/WRITE_ONCE() should also be used with anything
74  * that may change.
75  */
76 static uint64_t host_page_size;
77 static uint64_t guest_page_size;
78 static uint64_t guest_num_pages;
79 static uint64_t random_array[TEST_PAGES_PER_LOOP];
80 static uint64_t iteration;
81 
82 /*
83  * Guest physical memory offset of the testing memory slot.
84  * This will be set to the topmost valid physical address minus
85  * the test memory size.
86  */
87 static uint64_t guest_test_phys_mem;
88 
89 /*
90  * Guest virtual memory offset of the testing memory slot.
91  * Must not conflict with identity mapped test code.
92  */
93 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
94 
95 /*
96  * Continuously write to the first 8 bytes of a random pages within
97  * the testing memory region.
98  */
99 static void guest_code(void)
100 {
101 	uint64_t addr;
102 	int i;
103 
104 	/*
105 	 * On s390x, all pages of a 1M segment are initially marked as dirty
106 	 * when a page of the segment is written to for the very first time.
107 	 * To compensate this specialty in this test, we need to touch all
108 	 * pages during the first iteration.
109 	 */
110 	for (i = 0; i < guest_num_pages; i++) {
111 		addr = guest_test_virt_mem + i * guest_page_size;
112 		*(uint64_t *)addr = READ_ONCE(iteration);
113 	}
114 
115 	while (true) {
116 		for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
117 			addr = guest_test_virt_mem;
118 			addr += (READ_ONCE(random_array[i]) % guest_num_pages)
119 				* guest_page_size;
120 			addr = align_down(addr, host_page_size);
121 			*(uint64_t *)addr = READ_ONCE(iteration);
122 		}
123 
124 		/* Tell the host that we need more random numbers */
125 		GUEST_SYNC(1);
126 	}
127 }
128 
129 /* Host variables */
130 static bool host_quit;
131 
132 /* Points to the test VM memory region on which we track dirty logs */
133 static void *host_test_mem;
134 static uint64_t host_num_pages;
135 
136 /* For statistics only */
137 static uint64_t host_dirty_count;
138 static uint64_t host_clear_count;
139 static uint64_t host_track_next_count;
140 
141 /* Whether dirty ring reset is requested, or finished */
142 static sem_t sem_vcpu_stop;
143 static sem_t sem_vcpu_cont;
144 /*
145  * This is only set by main thread, and only cleared by vcpu thread.  It is
146  * used to request vcpu thread to stop at the next GUEST_SYNC, since GUEST_SYNC
147  * is the only place that we'll guarantee both "dirty bit" and "dirty data"
148  * will match.  E.g., SIG_IPI won't guarantee that if the vcpu is interrupted
149  * after setting dirty bit but before the data is written.
150  */
151 static atomic_t vcpu_sync_stop_requested;
152 /*
153  * This is updated by the vcpu thread to tell the host whether it's a
154  * ring-full event.  It should only be read until a sem_wait() of
155  * sem_vcpu_stop and before vcpu continues to run.
156  */
157 static bool dirty_ring_vcpu_ring_full;
158 /*
159  * This is only used for verifying the dirty pages.  Dirty ring has a very
160  * tricky case when the ring just got full, kvm will do userspace exit due to
161  * ring full.  When that happens, the very last PFN is set but actually the
162  * data is not changed (the guest WRITE is not really applied yet), because
163  * we found that the dirty ring is full, refused to continue the vcpu, and
164  * recorded the dirty gfn with the old contents.
165  *
166  * For this specific case, it's safe to skip checking this pfn for this
167  * bit, because it's a redundant bit, and when the write happens later the bit
168  * will be set again.  We use this variable to always keep track of the latest
169  * dirty gfn we've collected, so that if a mismatch of data found later in the
170  * verifying process, we let it pass.
171  */
172 static uint64_t dirty_ring_last_page;
173 
174 enum log_mode_t {
175 	/* Only use KVM_GET_DIRTY_LOG for logging */
176 	LOG_MODE_DIRTY_LOG = 0,
177 
178 	/* Use both KVM_[GET|CLEAR]_DIRTY_LOG for logging */
179 	LOG_MODE_CLEAR_LOG = 1,
180 
181 	/* Use dirty ring for logging */
182 	LOG_MODE_DIRTY_RING = 2,
183 
184 	LOG_MODE_NUM,
185 
186 	/* Run all supported modes */
187 	LOG_MODE_ALL = LOG_MODE_NUM,
188 };
189 
190 /* Mode of logging to test.  Default is to run all supported modes */
191 static enum log_mode_t host_log_mode_option = LOG_MODE_ALL;
192 /* Logging mode for current run */
193 static enum log_mode_t host_log_mode;
194 static pthread_t vcpu_thread;
195 static uint32_t test_dirty_ring_count = TEST_DIRTY_RING_COUNT;
196 
197 static void vcpu_kick(void)
198 {
199 	pthread_kill(vcpu_thread, SIG_IPI);
200 }
201 
202 /*
203  * In our test we do signal tricks, let's use a better version of
204  * sem_wait to avoid signal interrupts
205  */
206 static void sem_wait_until(sem_t *sem)
207 {
208 	int ret;
209 
210 	do
211 		ret = sem_wait(sem);
212 	while (ret == -1 && errno == EINTR);
213 }
214 
215 static bool clear_log_supported(void)
216 {
217 	return kvm_has_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
218 }
219 
220 static void clear_log_create_vm_done(struct kvm_vm *vm)
221 {
222 	u64 manual_caps;
223 
224 	manual_caps = kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
225 	TEST_ASSERT(manual_caps, "MANUAL_CAPS is zero!");
226 	manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE |
227 			KVM_DIRTY_LOG_INITIALLY_SET);
228 	vm_enable_cap(vm, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2, manual_caps);
229 }
230 
231 static void dirty_log_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
232 					  void *bitmap, uint32_t num_pages,
233 					  uint32_t *unused)
234 {
235 	kvm_vm_get_dirty_log(vcpu->vm, slot, bitmap);
236 }
237 
238 static void clear_log_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
239 					  void *bitmap, uint32_t num_pages,
240 					  uint32_t *unused)
241 {
242 	kvm_vm_get_dirty_log(vcpu->vm, slot, bitmap);
243 	kvm_vm_clear_dirty_log(vcpu->vm, slot, bitmap, 0, num_pages);
244 }
245 
246 /* Should only be called after a GUEST_SYNC */
247 static void vcpu_handle_sync_stop(void)
248 {
249 	if (atomic_read(&vcpu_sync_stop_requested)) {
250 		/* It means main thread is sleeping waiting */
251 		atomic_set(&vcpu_sync_stop_requested, false);
252 		sem_post(&sem_vcpu_stop);
253 		sem_wait_until(&sem_vcpu_cont);
254 	}
255 }
256 
257 static void default_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
258 {
259 	struct kvm_run *run = vcpu->run;
260 
261 	TEST_ASSERT(ret == 0 || (ret == -1 && err == EINTR),
262 		    "vcpu run failed: errno=%d", err);
263 
264 	TEST_ASSERT(get_ucall(vcpu, NULL) == UCALL_SYNC,
265 		    "Invalid guest sync status: exit_reason=%s\n",
266 		    exit_reason_str(run->exit_reason));
267 
268 	vcpu_handle_sync_stop();
269 }
270 
271 static bool dirty_ring_supported(void)
272 {
273 	return (kvm_has_cap(KVM_CAP_DIRTY_LOG_RING) ||
274 		kvm_has_cap(KVM_CAP_DIRTY_LOG_RING_ACQ_REL));
275 }
276 
277 static void dirty_ring_create_vm_done(struct kvm_vm *vm)
278 {
279 	uint64_t pages;
280 	uint32_t limit;
281 
282 	/*
283 	 * We rely on vcpu exit due to full dirty ring state. Adjust
284 	 * the ring buffer size to ensure we're able to reach the
285 	 * full dirty ring state.
286 	 */
287 	pages = (1ul << (DIRTY_MEM_BITS - vm->page_shift)) + 3;
288 	pages = vm_adjust_num_guest_pages(vm->mode, pages);
289 	if (vm->page_size < getpagesize())
290 		pages = vm_num_host_pages(vm->mode, pages);
291 
292 	limit = 1 << (31 - __builtin_clz(pages));
293 	test_dirty_ring_count = 1 << (31 - __builtin_clz(test_dirty_ring_count));
294 	test_dirty_ring_count = min(limit, test_dirty_ring_count);
295 	pr_info("dirty ring count: 0x%x\n", test_dirty_ring_count);
296 
297 	/*
298 	 * Switch to dirty ring mode after VM creation but before any
299 	 * of the vcpu creation.
300 	 */
301 	vm_enable_dirty_ring(vm, test_dirty_ring_count *
302 			     sizeof(struct kvm_dirty_gfn));
303 }
304 
305 static inline bool dirty_gfn_is_dirtied(struct kvm_dirty_gfn *gfn)
306 {
307 	return smp_load_acquire(&gfn->flags) == KVM_DIRTY_GFN_F_DIRTY;
308 }
309 
310 static inline void dirty_gfn_set_collected(struct kvm_dirty_gfn *gfn)
311 {
312 	smp_store_release(&gfn->flags, KVM_DIRTY_GFN_F_RESET);
313 }
314 
315 static uint32_t dirty_ring_collect_one(struct kvm_dirty_gfn *dirty_gfns,
316 				       int slot, void *bitmap,
317 				       uint32_t num_pages, uint32_t *fetch_index)
318 {
319 	struct kvm_dirty_gfn *cur;
320 	uint32_t count = 0;
321 
322 	while (true) {
323 		cur = &dirty_gfns[*fetch_index % test_dirty_ring_count];
324 		if (!dirty_gfn_is_dirtied(cur))
325 			break;
326 		TEST_ASSERT(cur->slot == slot, "Slot number didn't match: "
327 			    "%u != %u", cur->slot, slot);
328 		TEST_ASSERT(cur->offset < num_pages, "Offset overflow: "
329 			    "0x%llx >= 0x%x", cur->offset, num_pages);
330 		//pr_info("fetch 0x%x page %llu\n", *fetch_index, cur->offset);
331 		__set_bit_le(cur->offset, bitmap);
332 		dirty_ring_last_page = cur->offset;
333 		dirty_gfn_set_collected(cur);
334 		(*fetch_index)++;
335 		count++;
336 	}
337 
338 	return count;
339 }
340 
341 static void dirty_ring_wait_vcpu(void)
342 {
343 	/* This makes sure that hardware PML cache flushed */
344 	vcpu_kick();
345 	sem_wait_until(&sem_vcpu_stop);
346 }
347 
348 static void dirty_ring_continue_vcpu(void)
349 {
350 	pr_info("Notifying vcpu to continue\n");
351 	sem_post(&sem_vcpu_cont);
352 }
353 
354 static void dirty_ring_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
355 					   void *bitmap, uint32_t num_pages,
356 					   uint32_t *ring_buf_idx)
357 {
358 	uint32_t count = 0, cleared;
359 	bool continued_vcpu = false;
360 
361 	dirty_ring_wait_vcpu();
362 
363 	if (!dirty_ring_vcpu_ring_full) {
364 		/*
365 		 * This is not a ring-full event, it's safe to allow
366 		 * vcpu to continue
367 		 */
368 		dirty_ring_continue_vcpu();
369 		continued_vcpu = true;
370 	}
371 
372 	/* Only have one vcpu */
373 	count = dirty_ring_collect_one(vcpu_map_dirty_ring(vcpu),
374 				       slot, bitmap, num_pages,
375 				       ring_buf_idx);
376 
377 	cleared = kvm_vm_reset_dirty_ring(vcpu->vm);
378 
379 	/* Cleared pages should be the same as collected */
380 	TEST_ASSERT(cleared == count, "Reset dirty pages (%u) mismatch "
381 		    "with collected (%u)", cleared, count);
382 
383 	if (!continued_vcpu) {
384 		TEST_ASSERT(dirty_ring_vcpu_ring_full,
385 			    "Didn't continue vcpu even without ring full");
386 		dirty_ring_continue_vcpu();
387 	}
388 
389 	pr_info("Iteration %ld collected %u pages\n", iteration, count);
390 }
391 
392 static void dirty_ring_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
393 {
394 	struct kvm_run *run = vcpu->run;
395 
396 	/* A ucall-sync or ring-full event is allowed */
397 	if (get_ucall(vcpu, NULL) == UCALL_SYNC) {
398 		/* We should allow this to continue */
399 		;
400 	} else if (run->exit_reason == KVM_EXIT_DIRTY_RING_FULL ||
401 		   (ret == -1 && err == EINTR)) {
402 		/* Update the flag first before pause */
403 		WRITE_ONCE(dirty_ring_vcpu_ring_full,
404 			   run->exit_reason == KVM_EXIT_DIRTY_RING_FULL);
405 		sem_post(&sem_vcpu_stop);
406 		pr_info("vcpu stops because %s...\n",
407 			dirty_ring_vcpu_ring_full ?
408 			"dirty ring is full" : "vcpu is kicked out");
409 		sem_wait_until(&sem_vcpu_cont);
410 		pr_info("vcpu continues now.\n");
411 	} else {
412 		TEST_ASSERT(false, "Invalid guest sync status: "
413 			    "exit_reason=%s\n",
414 			    exit_reason_str(run->exit_reason));
415 	}
416 }
417 
418 static void dirty_ring_before_vcpu_join(void)
419 {
420 	/* Kick another round of vcpu just to make sure it will quit */
421 	sem_post(&sem_vcpu_cont);
422 }
423 
424 struct log_mode {
425 	const char *name;
426 	/* Return true if this mode is supported, otherwise false */
427 	bool (*supported)(void);
428 	/* Hook when the vm creation is done (before vcpu creation) */
429 	void (*create_vm_done)(struct kvm_vm *vm);
430 	/* Hook to collect the dirty pages into the bitmap provided */
431 	void (*collect_dirty_pages) (struct kvm_vcpu *vcpu, int slot,
432 				     void *bitmap, uint32_t num_pages,
433 				     uint32_t *ring_buf_idx);
434 	/* Hook to call when after each vcpu run */
435 	void (*after_vcpu_run)(struct kvm_vcpu *vcpu, int ret, int err);
436 	void (*before_vcpu_join) (void);
437 } log_modes[LOG_MODE_NUM] = {
438 	{
439 		.name = "dirty-log",
440 		.collect_dirty_pages = dirty_log_collect_dirty_pages,
441 		.after_vcpu_run = default_after_vcpu_run,
442 	},
443 	{
444 		.name = "clear-log",
445 		.supported = clear_log_supported,
446 		.create_vm_done = clear_log_create_vm_done,
447 		.collect_dirty_pages = clear_log_collect_dirty_pages,
448 		.after_vcpu_run = default_after_vcpu_run,
449 	},
450 	{
451 		.name = "dirty-ring",
452 		.supported = dirty_ring_supported,
453 		.create_vm_done = dirty_ring_create_vm_done,
454 		.collect_dirty_pages = dirty_ring_collect_dirty_pages,
455 		.before_vcpu_join = dirty_ring_before_vcpu_join,
456 		.after_vcpu_run = dirty_ring_after_vcpu_run,
457 	},
458 };
459 
460 /*
461  * We use this bitmap to track some pages that should have its dirty
462  * bit set in the _next_ iteration.  For example, if we detected the
463  * page value changed to current iteration but at the same time the
464  * page bit is cleared in the latest bitmap, then the system must
465  * report that write in the next get dirty log call.
466  */
467 static unsigned long *host_bmap_track;
468 
469 static void log_modes_dump(void)
470 {
471 	int i;
472 
473 	printf("all");
474 	for (i = 0; i < LOG_MODE_NUM; i++)
475 		printf(", %s", log_modes[i].name);
476 	printf("\n");
477 }
478 
479 static bool log_mode_supported(void)
480 {
481 	struct log_mode *mode = &log_modes[host_log_mode];
482 
483 	if (mode->supported)
484 		return mode->supported();
485 
486 	return true;
487 }
488 
489 static void log_mode_create_vm_done(struct kvm_vm *vm)
490 {
491 	struct log_mode *mode = &log_modes[host_log_mode];
492 
493 	if (mode->create_vm_done)
494 		mode->create_vm_done(vm);
495 }
496 
497 static void log_mode_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
498 					 void *bitmap, uint32_t num_pages,
499 					 uint32_t *ring_buf_idx)
500 {
501 	struct log_mode *mode = &log_modes[host_log_mode];
502 
503 	TEST_ASSERT(mode->collect_dirty_pages != NULL,
504 		    "collect_dirty_pages() is required for any log mode!");
505 	mode->collect_dirty_pages(vcpu, slot, bitmap, num_pages, ring_buf_idx);
506 }
507 
508 static void log_mode_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
509 {
510 	struct log_mode *mode = &log_modes[host_log_mode];
511 
512 	if (mode->after_vcpu_run)
513 		mode->after_vcpu_run(vcpu, ret, err);
514 }
515 
516 static void log_mode_before_vcpu_join(void)
517 {
518 	struct log_mode *mode = &log_modes[host_log_mode];
519 
520 	if (mode->before_vcpu_join)
521 		mode->before_vcpu_join();
522 }
523 
524 static void generate_random_array(uint64_t *guest_array, uint64_t size)
525 {
526 	uint64_t i;
527 
528 	for (i = 0; i < size; i++)
529 		guest_array[i] = random();
530 }
531 
532 static void *vcpu_worker(void *data)
533 {
534 	int ret;
535 	struct kvm_vcpu *vcpu = data;
536 	struct kvm_vm *vm = vcpu->vm;
537 	uint64_t *guest_array;
538 	uint64_t pages_count = 0;
539 	struct kvm_signal_mask *sigmask = alloca(offsetof(struct kvm_signal_mask, sigset)
540 						 + sizeof(sigset_t));
541 	sigset_t *sigset = (sigset_t *) &sigmask->sigset;
542 
543 	/*
544 	 * SIG_IPI is unblocked atomically while in KVM_RUN.  It causes the
545 	 * ioctl to return with -EINTR, but it is still pending and we need
546 	 * to accept it with the sigwait.
547 	 */
548 	sigmask->len = 8;
549 	pthread_sigmask(0, NULL, sigset);
550 	sigdelset(sigset, SIG_IPI);
551 	vcpu_ioctl(vcpu, KVM_SET_SIGNAL_MASK, sigmask);
552 
553 	sigemptyset(sigset);
554 	sigaddset(sigset, SIG_IPI);
555 
556 	guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
557 
558 	while (!READ_ONCE(host_quit)) {
559 		/* Clear any existing kick signals */
560 		generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
561 		pages_count += TEST_PAGES_PER_LOOP;
562 		/* Let the guest dirty the random pages */
563 		ret = __vcpu_run(vcpu);
564 		if (ret == -1 && errno == EINTR) {
565 			int sig = -1;
566 			sigwait(sigset, &sig);
567 			assert(sig == SIG_IPI);
568 		}
569 		log_mode_after_vcpu_run(vcpu, ret, errno);
570 	}
571 
572 	pr_info("Dirtied %"PRIu64" pages\n", pages_count);
573 
574 	return NULL;
575 }
576 
577 static void vm_dirty_log_verify(enum vm_guest_mode mode, unsigned long *bmap)
578 {
579 	uint64_t step = vm_num_host_pages(mode, 1);
580 	uint64_t page;
581 	uint64_t *value_ptr;
582 	uint64_t min_iter = 0;
583 
584 	for (page = 0; page < host_num_pages; page += step) {
585 		value_ptr = host_test_mem + page * host_page_size;
586 
587 		/* If this is a special page that we were tracking... */
588 		if (__test_and_clear_bit_le(page, host_bmap_track)) {
589 			host_track_next_count++;
590 			TEST_ASSERT(test_bit_le(page, bmap),
591 				    "Page %"PRIu64" should have its dirty bit "
592 				    "set in this iteration but it is missing",
593 				    page);
594 		}
595 
596 		if (__test_and_clear_bit_le(page, bmap)) {
597 			bool matched;
598 
599 			host_dirty_count++;
600 
601 			/*
602 			 * If the bit is set, the value written onto
603 			 * the corresponding page should be either the
604 			 * previous iteration number or the current one.
605 			 */
606 			matched = (*value_ptr == iteration ||
607 				   *value_ptr == iteration - 1);
608 
609 			if (host_log_mode == LOG_MODE_DIRTY_RING && !matched) {
610 				if (*value_ptr == iteration - 2 && min_iter <= iteration - 2) {
611 					/*
612 					 * Short answer: this case is special
613 					 * only for dirty ring test where the
614 					 * page is the last page before a kvm
615 					 * dirty ring full in iteration N-2.
616 					 *
617 					 * Long answer: Assuming ring size R,
618 					 * one possible condition is:
619 					 *
620 					 *      main thr       vcpu thr
621 					 *      --------       --------
622 					 *    iter=1
623 					 *                   write 1 to page 0~(R-1)
624 					 *                   full, vmexit
625 					 *    collect 0~(R-1)
626 					 *    kick vcpu
627 					 *                   write 1 to (R-1)~(2R-2)
628 					 *                   full, vmexit
629 					 *    iter=2
630 					 *    collect (R-1)~(2R-2)
631 					 *    kick vcpu
632 					 *                   write 1 to (2R-2)
633 					 *                   (NOTE!!! "1" cached in cpu reg)
634 					 *                   write 2 to (2R-1)~(3R-3)
635 					 *                   full, vmexit
636 					 *    iter=3
637 					 *    collect (2R-2)~(3R-3)
638 					 *    (here if we read value on page
639 					 *     "2R-2" is 1, while iter=3!!!)
640 					 *
641 					 * This however can only happen once per iteration.
642 					 */
643 					min_iter = iteration - 1;
644 					continue;
645 				} else if (page == dirty_ring_last_page) {
646 					/*
647 					 * Please refer to comments in
648 					 * dirty_ring_last_page.
649 					 */
650 					continue;
651 				}
652 			}
653 
654 			TEST_ASSERT(matched,
655 				    "Set page %"PRIu64" value %"PRIu64
656 				    " incorrect (iteration=%"PRIu64")",
657 				    page, *value_ptr, iteration);
658 		} else {
659 			host_clear_count++;
660 			/*
661 			 * If cleared, the value written can be any
662 			 * value smaller or equals to the iteration
663 			 * number.  Note that the value can be exactly
664 			 * (iteration-1) if that write can happen
665 			 * like this:
666 			 *
667 			 * (1) increase loop count to "iteration-1"
668 			 * (2) write to page P happens (with value
669 			 *     "iteration-1")
670 			 * (3) get dirty log for "iteration-1"; we'll
671 			 *     see that page P bit is set (dirtied),
672 			 *     and not set the bit in host_bmap_track
673 			 * (4) increase loop count to "iteration"
674 			 *     (which is current iteration)
675 			 * (5) get dirty log for current iteration,
676 			 *     we'll see that page P is cleared, with
677 			 *     value "iteration-1".
678 			 */
679 			TEST_ASSERT(*value_ptr <= iteration,
680 				    "Clear page %"PRIu64" value %"PRIu64
681 				    " incorrect (iteration=%"PRIu64")",
682 				    page, *value_ptr, iteration);
683 			if (*value_ptr == iteration) {
684 				/*
685 				 * This page is _just_ modified; it
686 				 * should report its dirtyness in the
687 				 * next run
688 				 */
689 				__set_bit_le(page, host_bmap_track);
690 			}
691 		}
692 	}
693 }
694 
695 static struct kvm_vm *create_vm(enum vm_guest_mode mode, struct kvm_vcpu **vcpu,
696 				uint64_t extra_mem_pages, void *guest_code)
697 {
698 	struct kvm_vm *vm;
699 
700 	pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
701 
702 	vm = __vm_create(mode, 1, extra_mem_pages);
703 
704 	log_mode_create_vm_done(vm);
705 	*vcpu = vm_vcpu_add(vm, 0, guest_code);
706 	return vm;
707 }
708 
709 struct test_params {
710 	unsigned long iterations;
711 	unsigned long interval;
712 	uint64_t phys_offset;
713 };
714 
715 static void run_test(enum vm_guest_mode mode, void *arg)
716 {
717 	struct test_params *p = arg;
718 	struct kvm_vcpu *vcpu;
719 	struct kvm_vm *vm;
720 	unsigned long *bmap;
721 	uint32_t ring_buf_idx = 0;
722 
723 	if (!log_mode_supported()) {
724 		print_skip("Log mode '%s' not supported",
725 			   log_modes[host_log_mode].name);
726 		return;
727 	}
728 
729 	/*
730 	 * We reserve page table for 2 times of extra dirty mem which
731 	 * will definitely cover the original (1G+) test range.  Here
732 	 * we do the calculation with 4K page size which is the
733 	 * smallest so the page number will be enough for all archs
734 	 * (e.g., 64K page size guest will need even less memory for
735 	 * page tables).
736 	 */
737 	vm = create_vm(mode, &vcpu,
738 		       2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K), guest_code);
739 
740 	guest_page_size = vm->page_size;
741 	/*
742 	 * A little more than 1G of guest page sized pages.  Cover the
743 	 * case where the size is not aligned to 64 pages.
744 	 */
745 	guest_num_pages = (1ul << (DIRTY_MEM_BITS - vm->page_shift)) + 3;
746 	guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
747 
748 	host_page_size = getpagesize();
749 	host_num_pages = vm_num_host_pages(mode, guest_num_pages);
750 
751 	if (!p->phys_offset) {
752 		guest_test_phys_mem = (vm->max_gfn - guest_num_pages) *
753 				      guest_page_size;
754 		guest_test_phys_mem = align_down(guest_test_phys_mem, host_page_size);
755 	} else {
756 		guest_test_phys_mem = p->phys_offset;
757 	}
758 
759 #ifdef __s390x__
760 	/* Align to 1M (segment size) */
761 	guest_test_phys_mem = align_down(guest_test_phys_mem, 1 << 20);
762 #endif
763 
764 	pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
765 
766 	bmap = bitmap_zalloc(host_num_pages);
767 	host_bmap_track = bitmap_zalloc(host_num_pages);
768 
769 	/* Add an extra memory slot for testing dirty logging */
770 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
771 				    guest_test_phys_mem,
772 				    TEST_MEM_SLOT_INDEX,
773 				    guest_num_pages,
774 				    KVM_MEM_LOG_DIRTY_PAGES);
775 
776 	/* Do mapping for the dirty track memory slot */
777 	virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages);
778 
779 	/* Cache the HVA pointer of the region */
780 	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
781 
782 	/* Export the shared variables to the guest */
783 	sync_global_to_guest(vm, host_page_size);
784 	sync_global_to_guest(vm, guest_page_size);
785 	sync_global_to_guest(vm, guest_test_virt_mem);
786 	sync_global_to_guest(vm, guest_num_pages);
787 
788 	/* Start the iterations */
789 	iteration = 1;
790 	sync_global_to_guest(vm, iteration);
791 	host_quit = false;
792 	host_dirty_count = 0;
793 	host_clear_count = 0;
794 	host_track_next_count = 0;
795 	WRITE_ONCE(dirty_ring_vcpu_ring_full, false);
796 
797 	pthread_create(&vcpu_thread, NULL, vcpu_worker, vcpu);
798 
799 	while (iteration < p->iterations) {
800 		/* Give the vcpu thread some time to dirty some pages */
801 		usleep(p->interval * 1000);
802 		log_mode_collect_dirty_pages(vcpu, TEST_MEM_SLOT_INDEX,
803 					     bmap, host_num_pages,
804 					     &ring_buf_idx);
805 
806 		/*
807 		 * See vcpu_sync_stop_requested definition for details on why
808 		 * we need to stop vcpu when verify data.
809 		 */
810 		atomic_set(&vcpu_sync_stop_requested, true);
811 		sem_wait_until(&sem_vcpu_stop);
812 		/*
813 		 * NOTE: for dirty ring, it's possible that we didn't stop at
814 		 * GUEST_SYNC but instead we stopped because ring is full;
815 		 * that's okay too because ring full means we're only missing
816 		 * the flush of the last page, and since we handle the last
817 		 * page specially verification will succeed anyway.
818 		 */
819 		assert(host_log_mode == LOG_MODE_DIRTY_RING ||
820 		       atomic_read(&vcpu_sync_stop_requested) == false);
821 		vm_dirty_log_verify(mode, bmap);
822 		sem_post(&sem_vcpu_cont);
823 
824 		iteration++;
825 		sync_global_to_guest(vm, iteration);
826 	}
827 
828 	/* Tell the vcpu thread to quit */
829 	host_quit = true;
830 	log_mode_before_vcpu_join();
831 	pthread_join(vcpu_thread, NULL);
832 
833 	pr_info("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
834 		"track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
835 		host_track_next_count);
836 
837 	free(bmap);
838 	free(host_bmap_track);
839 	kvm_vm_free(vm);
840 }
841 
842 static void help(char *name)
843 {
844 	puts("");
845 	printf("usage: %s [-h] [-i iterations] [-I interval] "
846 	       "[-p offset] [-m mode]\n", name);
847 	puts("");
848 	printf(" -c: hint to dirty ring size, in number of entries\n");
849 	printf("     (only useful for dirty-ring test; default: %"PRIu32")\n",
850 	       TEST_DIRTY_RING_COUNT);
851 	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
852 	       TEST_HOST_LOOP_N);
853 	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
854 	       TEST_HOST_LOOP_INTERVAL);
855 	printf(" -p: specify guest physical test memory offset\n"
856 	       "     Warning: a low offset can conflict with the loaded test code.\n");
857 	printf(" -M: specify the host logging mode "
858 	       "(default: run all log modes).  Supported modes: \n\t");
859 	log_modes_dump();
860 	guest_modes_help();
861 	puts("");
862 	exit(0);
863 }
864 
865 int main(int argc, char *argv[])
866 {
867 	struct test_params p = {
868 		.iterations = TEST_HOST_LOOP_N,
869 		.interval = TEST_HOST_LOOP_INTERVAL,
870 	};
871 	int opt, i;
872 	sigset_t sigset;
873 
874 	sem_init(&sem_vcpu_stop, 0, 0);
875 	sem_init(&sem_vcpu_cont, 0, 0);
876 
877 	guest_modes_append_default();
878 
879 	while ((opt = getopt(argc, argv, "c:hi:I:p:m:M:")) != -1) {
880 		switch (opt) {
881 		case 'c':
882 			test_dirty_ring_count = strtol(optarg, NULL, 10);
883 			break;
884 		case 'i':
885 			p.iterations = strtol(optarg, NULL, 10);
886 			break;
887 		case 'I':
888 			p.interval = strtol(optarg, NULL, 10);
889 			break;
890 		case 'p':
891 			p.phys_offset = strtoull(optarg, NULL, 0);
892 			break;
893 		case 'm':
894 			guest_modes_cmdline(optarg);
895 			break;
896 		case 'M':
897 			if (!strcmp(optarg, "all")) {
898 				host_log_mode_option = LOG_MODE_ALL;
899 				break;
900 			}
901 			for (i = 0; i < LOG_MODE_NUM; i++) {
902 				if (!strcmp(optarg, log_modes[i].name)) {
903 					pr_info("Setting log mode to: '%s'\n",
904 						optarg);
905 					host_log_mode_option = i;
906 					break;
907 				}
908 			}
909 			if (i == LOG_MODE_NUM) {
910 				printf("Log mode '%s' invalid. Please choose "
911 				       "from: ", optarg);
912 				log_modes_dump();
913 				exit(1);
914 			}
915 			break;
916 		case 'h':
917 		default:
918 			help(argv[0]);
919 			break;
920 		}
921 	}
922 
923 	TEST_ASSERT(p.iterations > 2, "Iterations must be greater than two");
924 	TEST_ASSERT(p.interval > 0, "Interval must be greater than zero");
925 
926 	pr_info("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
927 		p.iterations, p.interval);
928 
929 	srandom(time(0));
930 
931 	/* Ensure that vCPU threads start with SIG_IPI blocked.  */
932 	sigemptyset(&sigset);
933 	sigaddset(&sigset, SIG_IPI);
934 	pthread_sigmask(SIG_BLOCK, &sigset, NULL);
935 
936 	if (host_log_mode_option == LOG_MODE_ALL) {
937 		/* Run each log mode */
938 		for (i = 0; i < LOG_MODE_NUM; i++) {
939 			pr_info("Testing Log Mode '%s'\n", log_modes[i].name);
940 			host_log_mode = i;
941 			for_each_guest_mode(run_test, &p);
942 		}
943 	} else {
944 		host_log_mode = host_log_mode_option;
945 		for_each_guest_mode(run_test, &p);
946 	}
947 
948 	return 0;
949 }
950