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 	/*
380 	 * Cleared pages should be the same as collected, as KVM is supposed to
381 	 * clear only the entries that have been harvested.
382 	 */
383 	TEST_ASSERT(cleared == count, "Reset dirty pages (%u) mismatch "
384 		    "with collected (%u)", cleared, count);
385 
386 	if (!continued_vcpu) {
387 		TEST_ASSERT(dirty_ring_vcpu_ring_full,
388 			    "Didn't continue vcpu even without ring full");
389 		dirty_ring_continue_vcpu();
390 	}
391 
392 	pr_info("Iteration %ld collected %u pages\n", iteration, count);
393 }
394 
395 static void dirty_ring_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
396 {
397 	struct kvm_run *run = vcpu->run;
398 
399 	/* A ucall-sync or ring-full event is allowed */
400 	if (get_ucall(vcpu, NULL) == UCALL_SYNC) {
401 		/* We should allow this to continue */
402 		;
403 	} else if (run->exit_reason == KVM_EXIT_DIRTY_RING_FULL ||
404 		   (ret == -1 && err == EINTR)) {
405 		/* Update the flag first before pause */
406 		WRITE_ONCE(dirty_ring_vcpu_ring_full,
407 			   run->exit_reason == KVM_EXIT_DIRTY_RING_FULL);
408 		sem_post(&sem_vcpu_stop);
409 		pr_info("vcpu stops because %s...\n",
410 			dirty_ring_vcpu_ring_full ?
411 			"dirty ring is full" : "vcpu is kicked out");
412 		sem_wait_until(&sem_vcpu_cont);
413 		pr_info("vcpu continues now.\n");
414 	} else {
415 		TEST_ASSERT(false, "Invalid guest sync status: "
416 			    "exit_reason=%s\n",
417 			    exit_reason_str(run->exit_reason));
418 	}
419 }
420 
421 struct log_mode {
422 	const char *name;
423 	/* Return true if this mode is supported, otherwise false */
424 	bool (*supported)(void);
425 	/* Hook when the vm creation is done (before vcpu creation) */
426 	void (*create_vm_done)(struct kvm_vm *vm);
427 	/* Hook to collect the dirty pages into the bitmap provided */
428 	void (*collect_dirty_pages) (struct kvm_vcpu *vcpu, int slot,
429 				     void *bitmap, uint32_t num_pages,
430 				     uint32_t *ring_buf_idx);
431 	/* Hook to call when after each vcpu run */
432 	void (*after_vcpu_run)(struct kvm_vcpu *vcpu, int ret, int err);
433 } log_modes[LOG_MODE_NUM] = {
434 	{
435 		.name = "dirty-log",
436 		.collect_dirty_pages = dirty_log_collect_dirty_pages,
437 		.after_vcpu_run = default_after_vcpu_run,
438 	},
439 	{
440 		.name = "clear-log",
441 		.supported = clear_log_supported,
442 		.create_vm_done = clear_log_create_vm_done,
443 		.collect_dirty_pages = clear_log_collect_dirty_pages,
444 		.after_vcpu_run = default_after_vcpu_run,
445 	},
446 	{
447 		.name = "dirty-ring",
448 		.supported = dirty_ring_supported,
449 		.create_vm_done = dirty_ring_create_vm_done,
450 		.collect_dirty_pages = dirty_ring_collect_dirty_pages,
451 		.after_vcpu_run = dirty_ring_after_vcpu_run,
452 	},
453 };
454 
455 /*
456  * We use this bitmap to track some pages that should have its dirty
457  * bit set in the _next_ iteration.  For example, if we detected the
458  * page value changed to current iteration but at the same time the
459  * page bit is cleared in the latest bitmap, then the system must
460  * report that write in the next get dirty log call.
461  */
462 static unsigned long *host_bmap_track;
463 
464 static void log_modes_dump(void)
465 {
466 	int i;
467 
468 	printf("all");
469 	for (i = 0; i < LOG_MODE_NUM; i++)
470 		printf(", %s", log_modes[i].name);
471 	printf("\n");
472 }
473 
474 static bool log_mode_supported(void)
475 {
476 	struct log_mode *mode = &log_modes[host_log_mode];
477 
478 	if (mode->supported)
479 		return mode->supported();
480 
481 	return true;
482 }
483 
484 static void log_mode_create_vm_done(struct kvm_vm *vm)
485 {
486 	struct log_mode *mode = &log_modes[host_log_mode];
487 
488 	if (mode->create_vm_done)
489 		mode->create_vm_done(vm);
490 }
491 
492 static void log_mode_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
493 					 void *bitmap, uint32_t num_pages,
494 					 uint32_t *ring_buf_idx)
495 {
496 	struct log_mode *mode = &log_modes[host_log_mode];
497 
498 	TEST_ASSERT(mode->collect_dirty_pages != NULL,
499 		    "collect_dirty_pages() is required for any log mode!");
500 	mode->collect_dirty_pages(vcpu, slot, bitmap, num_pages, ring_buf_idx);
501 }
502 
503 static void log_mode_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
504 {
505 	struct log_mode *mode = &log_modes[host_log_mode];
506 
507 	if (mode->after_vcpu_run)
508 		mode->after_vcpu_run(vcpu, ret, err);
509 }
510 
511 static void generate_random_array(uint64_t *guest_array, uint64_t size)
512 {
513 	uint64_t i;
514 
515 	for (i = 0; i < size; i++)
516 		guest_array[i] = random();
517 }
518 
519 static void *vcpu_worker(void *data)
520 {
521 	int ret;
522 	struct kvm_vcpu *vcpu = data;
523 	struct kvm_vm *vm = vcpu->vm;
524 	uint64_t *guest_array;
525 	uint64_t pages_count = 0;
526 	struct kvm_signal_mask *sigmask = alloca(offsetof(struct kvm_signal_mask, sigset)
527 						 + sizeof(sigset_t));
528 	sigset_t *sigset = (sigset_t *) &sigmask->sigset;
529 
530 	/*
531 	 * SIG_IPI is unblocked atomically while in KVM_RUN.  It causes the
532 	 * ioctl to return with -EINTR, but it is still pending and we need
533 	 * to accept it with the sigwait.
534 	 */
535 	sigmask->len = 8;
536 	pthread_sigmask(0, NULL, sigset);
537 	sigdelset(sigset, SIG_IPI);
538 	vcpu_ioctl(vcpu, KVM_SET_SIGNAL_MASK, sigmask);
539 
540 	sigemptyset(sigset);
541 	sigaddset(sigset, SIG_IPI);
542 
543 	guest_array = addr_gva2hva(vm, (vm_vaddr_t)random_array);
544 
545 	while (!READ_ONCE(host_quit)) {
546 		/* Clear any existing kick signals */
547 		generate_random_array(guest_array, TEST_PAGES_PER_LOOP);
548 		pages_count += TEST_PAGES_PER_LOOP;
549 		/* Let the guest dirty the random pages */
550 		ret = __vcpu_run(vcpu);
551 		if (ret == -1 && errno == EINTR) {
552 			int sig = -1;
553 			sigwait(sigset, &sig);
554 			assert(sig == SIG_IPI);
555 		}
556 		log_mode_after_vcpu_run(vcpu, ret, errno);
557 	}
558 
559 	pr_info("Dirtied %"PRIu64" pages\n", pages_count);
560 
561 	return NULL;
562 }
563 
564 static void vm_dirty_log_verify(enum vm_guest_mode mode, unsigned long *bmap)
565 {
566 	uint64_t step = vm_num_host_pages(mode, 1);
567 	uint64_t page;
568 	uint64_t *value_ptr;
569 	uint64_t min_iter = 0;
570 
571 	for (page = 0; page < host_num_pages; page += step) {
572 		value_ptr = host_test_mem + page * host_page_size;
573 
574 		/* If this is a special page that we were tracking... */
575 		if (__test_and_clear_bit_le(page, host_bmap_track)) {
576 			host_track_next_count++;
577 			TEST_ASSERT(test_bit_le(page, bmap),
578 				    "Page %"PRIu64" should have its dirty bit "
579 				    "set in this iteration but it is missing",
580 				    page);
581 		}
582 
583 		if (__test_and_clear_bit_le(page, bmap)) {
584 			bool matched;
585 
586 			host_dirty_count++;
587 
588 			/*
589 			 * If the bit is set, the value written onto
590 			 * the corresponding page should be either the
591 			 * previous iteration number or the current one.
592 			 */
593 			matched = (*value_ptr == iteration ||
594 				   *value_ptr == iteration - 1);
595 
596 			if (host_log_mode == LOG_MODE_DIRTY_RING && !matched) {
597 				if (*value_ptr == iteration - 2 && min_iter <= iteration - 2) {
598 					/*
599 					 * Short answer: this case is special
600 					 * only for dirty ring test where the
601 					 * page is the last page before a kvm
602 					 * dirty ring full in iteration N-2.
603 					 *
604 					 * Long answer: Assuming ring size R,
605 					 * one possible condition is:
606 					 *
607 					 *      main thr       vcpu thr
608 					 *      --------       --------
609 					 *    iter=1
610 					 *                   write 1 to page 0~(R-1)
611 					 *                   full, vmexit
612 					 *    collect 0~(R-1)
613 					 *    kick vcpu
614 					 *                   write 1 to (R-1)~(2R-2)
615 					 *                   full, vmexit
616 					 *    iter=2
617 					 *    collect (R-1)~(2R-2)
618 					 *    kick vcpu
619 					 *                   write 1 to (2R-2)
620 					 *                   (NOTE!!! "1" cached in cpu reg)
621 					 *                   write 2 to (2R-1)~(3R-3)
622 					 *                   full, vmexit
623 					 *    iter=3
624 					 *    collect (2R-2)~(3R-3)
625 					 *    (here if we read value on page
626 					 *     "2R-2" is 1, while iter=3!!!)
627 					 *
628 					 * This however can only happen once per iteration.
629 					 */
630 					min_iter = iteration - 1;
631 					continue;
632 				} else if (page == dirty_ring_last_page) {
633 					/*
634 					 * Please refer to comments in
635 					 * dirty_ring_last_page.
636 					 */
637 					continue;
638 				}
639 			}
640 
641 			TEST_ASSERT(matched,
642 				    "Set page %"PRIu64" value %"PRIu64
643 				    " incorrect (iteration=%"PRIu64")",
644 				    page, *value_ptr, iteration);
645 		} else {
646 			host_clear_count++;
647 			/*
648 			 * If cleared, the value written can be any
649 			 * value smaller or equals to the iteration
650 			 * number.  Note that the value can be exactly
651 			 * (iteration-1) if that write can happen
652 			 * like this:
653 			 *
654 			 * (1) increase loop count to "iteration-1"
655 			 * (2) write to page P happens (with value
656 			 *     "iteration-1")
657 			 * (3) get dirty log for "iteration-1"; we'll
658 			 *     see that page P bit is set (dirtied),
659 			 *     and not set the bit in host_bmap_track
660 			 * (4) increase loop count to "iteration"
661 			 *     (which is current iteration)
662 			 * (5) get dirty log for current iteration,
663 			 *     we'll see that page P is cleared, with
664 			 *     value "iteration-1".
665 			 */
666 			TEST_ASSERT(*value_ptr <= iteration,
667 				    "Clear page %"PRIu64" value %"PRIu64
668 				    " incorrect (iteration=%"PRIu64")",
669 				    page, *value_ptr, iteration);
670 			if (*value_ptr == iteration) {
671 				/*
672 				 * This page is _just_ modified; it
673 				 * should report its dirtyness in the
674 				 * next run
675 				 */
676 				__set_bit_le(page, host_bmap_track);
677 			}
678 		}
679 	}
680 }
681 
682 static struct kvm_vm *create_vm(enum vm_guest_mode mode, struct kvm_vcpu **vcpu,
683 				uint64_t extra_mem_pages, void *guest_code)
684 {
685 	struct kvm_vm *vm;
686 
687 	pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
688 
689 	vm = __vm_create(mode, 1, extra_mem_pages);
690 
691 	log_mode_create_vm_done(vm);
692 	*vcpu = vm_vcpu_add(vm, 0, guest_code);
693 	return vm;
694 }
695 
696 struct test_params {
697 	unsigned long iterations;
698 	unsigned long interval;
699 	uint64_t phys_offset;
700 };
701 
702 static void run_test(enum vm_guest_mode mode, void *arg)
703 {
704 	struct test_params *p = arg;
705 	struct kvm_vcpu *vcpu;
706 	struct kvm_vm *vm;
707 	unsigned long *bmap;
708 	uint32_t ring_buf_idx = 0;
709 	int sem_val;
710 
711 	if (!log_mode_supported()) {
712 		print_skip("Log mode '%s' not supported",
713 			   log_modes[host_log_mode].name);
714 		return;
715 	}
716 
717 	/*
718 	 * We reserve page table for 2 times of extra dirty mem which
719 	 * will definitely cover the original (1G+) test range.  Here
720 	 * we do the calculation with 4K page size which is the
721 	 * smallest so the page number will be enough for all archs
722 	 * (e.g., 64K page size guest will need even less memory for
723 	 * page tables).
724 	 */
725 	vm = create_vm(mode, &vcpu,
726 		       2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K), guest_code);
727 
728 	guest_page_size = vm->page_size;
729 	/*
730 	 * A little more than 1G of guest page sized pages.  Cover the
731 	 * case where the size is not aligned to 64 pages.
732 	 */
733 	guest_num_pages = (1ul << (DIRTY_MEM_BITS - vm->page_shift)) + 3;
734 	guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
735 
736 	host_page_size = getpagesize();
737 	host_num_pages = vm_num_host_pages(mode, guest_num_pages);
738 
739 	if (!p->phys_offset) {
740 		guest_test_phys_mem = (vm->max_gfn - guest_num_pages) *
741 				      guest_page_size;
742 		guest_test_phys_mem = align_down(guest_test_phys_mem, host_page_size);
743 	} else {
744 		guest_test_phys_mem = p->phys_offset;
745 	}
746 
747 #ifdef __s390x__
748 	/* Align to 1M (segment size) */
749 	guest_test_phys_mem = align_down(guest_test_phys_mem, 1 << 20);
750 #endif
751 
752 	pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
753 
754 	bmap = bitmap_zalloc(host_num_pages);
755 	host_bmap_track = bitmap_zalloc(host_num_pages);
756 
757 	/* Add an extra memory slot for testing dirty logging */
758 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
759 				    guest_test_phys_mem,
760 				    TEST_MEM_SLOT_INDEX,
761 				    guest_num_pages,
762 				    KVM_MEM_LOG_DIRTY_PAGES);
763 
764 	/* Do mapping for the dirty track memory slot */
765 	virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages);
766 
767 	/* Cache the HVA pointer of the region */
768 	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
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 	WRITE_ONCE(host_quit, false);
780 	host_dirty_count = 0;
781 	host_clear_count = 0;
782 	host_track_next_count = 0;
783 	WRITE_ONCE(dirty_ring_vcpu_ring_full, false);
784 
785 	/*
786 	 * Ensure the previous iteration didn't leave a dangling semaphore, i.e.
787 	 * that the main task and vCPU worker were synchronized and completed
788 	 * verification of all iterations.
789 	 */
790 	sem_getvalue(&sem_vcpu_stop, &sem_val);
791 	TEST_ASSERT_EQ(sem_val, 0);
792 	sem_getvalue(&sem_vcpu_cont, &sem_val);
793 	TEST_ASSERT_EQ(sem_val, 0);
794 
795 	pthread_create(&vcpu_thread, NULL, vcpu_worker, vcpu);
796 
797 	while (iteration < p->iterations) {
798 		/* Give the vcpu thread some time to dirty some pages */
799 		usleep(p->interval * 1000);
800 		log_mode_collect_dirty_pages(vcpu, TEST_MEM_SLOT_INDEX,
801 					     bmap, host_num_pages,
802 					     &ring_buf_idx);
803 
804 		/*
805 		 * See vcpu_sync_stop_requested definition for details on why
806 		 * we need to stop vcpu when verify data.
807 		 */
808 		atomic_set(&vcpu_sync_stop_requested, true);
809 		sem_wait_until(&sem_vcpu_stop);
810 		/*
811 		 * NOTE: for dirty ring, it's possible that we didn't stop at
812 		 * GUEST_SYNC but instead we stopped because ring is full;
813 		 * that's okay too because ring full means we're only missing
814 		 * the flush of the last page, and since we handle the last
815 		 * page specially verification will succeed anyway.
816 		 */
817 		assert(host_log_mode == LOG_MODE_DIRTY_RING ||
818 		       atomic_read(&vcpu_sync_stop_requested) == false);
819 		vm_dirty_log_verify(mode, bmap);
820 
821 		/*
822 		 * Set host_quit before sem_vcpu_cont in the final iteration to
823 		 * ensure that the vCPU worker doesn't resume the guest.  As
824 		 * above, the dirty ring test may stop and wait even when not
825 		 * explicitly request to do so, i.e. would hang waiting for a
826 		 * "continue" if it's allowed to resume the guest.
827 		 */
828 		if (++iteration == p->iterations)
829 			WRITE_ONCE(host_quit, true);
830 
831 		sem_post(&sem_vcpu_cont);
832 		sync_global_to_guest(vm, iteration);
833 	}
834 
835 	pthread_join(vcpu_thread, NULL);
836 
837 	pr_info("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
838 		"track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
839 		host_track_next_count);
840 
841 	free(bmap);
842 	free(host_bmap_track);
843 	kvm_vm_free(vm);
844 }
845 
846 static void help(char *name)
847 {
848 	puts("");
849 	printf("usage: %s [-h] [-i iterations] [-I interval] "
850 	       "[-p offset] [-m mode]\n", name);
851 	puts("");
852 	printf(" -c: hint to dirty ring size, in number of entries\n");
853 	printf("     (only useful for dirty-ring test; default: %"PRIu32")\n",
854 	       TEST_DIRTY_RING_COUNT);
855 	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
856 	       TEST_HOST_LOOP_N);
857 	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
858 	       TEST_HOST_LOOP_INTERVAL);
859 	printf(" -p: specify guest physical test memory offset\n"
860 	       "     Warning: a low offset can conflict with the loaded test code.\n");
861 	printf(" -M: specify the host logging mode "
862 	       "(default: run all log modes).  Supported modes: \n\t");
863 	log_modes_dump();
864 	guest_modes_help();
865 	puts("");
866 	exit(0);
867 }
868 
869 int main(int argc, char *argv[])
870 {
871 	struct test_params p = {
872 		.iterations = TEST_HOST_LOOP_N,
873 		.interval = TEST_HOST_LOOP_INTERVAL,
874 	};
875 	int opt, i;
876 	sigset_t sigset;
877 
878 	sem_init(&sem_vcpu_stop, 0, 0);
879 	sem_init(&sem_vcpu_cont, 0, 0);
880 
881 	guest_modes_append_default();
882 
883 	while ((opt = getopt(argc, argv, "c:hi:I:p:m:M:")) != -1) {
884 		switch (opt) {
885 		case 'c':
886 			test_dirty_ring_count = strtol(optarg, NULL, 10);
887 			break;
888 		case 'i':
889 			p.iterations = strtol(optarg, NULL, 10);
890 			break;
891 		case 'I':
892 			p.interval = strtol(optarg, NULL, 10);
893 			break;
894 		case 'p':
895 			p.phys_offset = strtoull(optarg, NULL, 0);
896 			break;
897 		case 'm':
898 			guest_modes_cmdline(optarg);
899 			break;
900 		case 'M':
901 			if (!strcmp(optarg, "all")) {
902 				host_log_mode_option = LOG_MODE_ALL;
903 				break;
904 			}
905 			for (i = 0; i < LOG_MODE_NUM; i++) {
906 				if (!strcmp(optarg, log_modes[i].name)) {
907 					pr_info("Setting log mode to: '%s'\n",
908 						optarg);
909 					host_log_mode_option = i;
910 					break;
911 				}
912 			}
913 			if (i == LOG_MODE_NUM) {
914 				printf("Log mode '%s' invalid. Please choose "
915 				       "from: ", optarg);
916 				log_modes_dump();
917 				exit(1);
918 			}
919 			break;
920 		case 'h':
921 		default:
922 			help(argv[0]);
923 			break;
924 		}
925 	}
926 
927 	TEST_ASSERT(p.iterations > 2, "Iterations must be greater than two");
928 	TEST_ASSERT(p.interval > 0, "Interval must be greater than zero");
929 
930 	pr_info("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
931 		p.iterations, p.interval);
932 
933 	srandom(time(0));
934 
935 	/* Ensure that vCPU threads start with SIG_IPI blocked.  */
936 	sigemptyset(&sigset);
937 	sigaddset(&sigset, SIG_IPI);
938 	pthread_sigmask(SIG_BLOCK, &sigset, NULL);
939 
940 	if (host_log_mode_option == LOG_MODE_ALL) {
941 		/* Run each log mode */
942 		for (i = 0; i < LOG_MODE_NUM; i++) {
943 			pr_info("Testing Log Mode '%s'\n", log_modes[i].name);
944 			host_log_mode = i;
945 			for_each_guest_mode(run_test, &p);
946 		}
947 	} else {
948 		host_log_mode = host_log_mode_option;
949 		for_each_guest_mode(run_test, &p);
950 	}
951 
952 	return 0;
953 }
954