xref: /linux/tools/testing/selftests/kvm/dirty_log_test.c (revision f4b0c4b508364fde023e4f7b9f23f7e38c663dfe)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * KVM dirty page logging test
4  *
5  * Copyright (C) 2018, Red Hat, Inc.
6  */
7 #include <stdio.h>
8 #include <stdlib.h>
9 #include <pthread.h>
10 #include <semaphore.h>
11 #include <sys/types.h>
12 #include <signal.h>
13 #include <errno.h>
14 #include <linux/bitmap.h>
15 #include <linux/bitops.h>
16 #include <linux/atomic.h>
17 #include <asm/barrier.h>
18 
19 #include "kvm_util.h"
20 #include "test_util.h"
21 #include "guest_modes.h"
22 #include "processor.h"
23 #include "ucall_common.h"
24 
25 #define DIRTY_MEM_BITS 30 /* 1G */
26 #define PAGE_SHIFT_4K  12
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 iteration;
78 
79 /*
80  * Guest physical memory offset of the testing memory slot.
81  * This will be set to the topmost valid physical address minus
82  * the test memory size.
83  */
84 static uint64_t guest_test_phys_mem;
85 
86 /*
87  * Guest virtual memory offset of the testing memory slot.
88  * Must not conflict with identity mapped test code.
89  */
90 static uint64_t guest_test_virt_mem = DEFAULT_GUEST_TEST_MEM;
91 
92 /*
93  * Continuously write to the first 8 bytes of a random pages within
94  * the testing memory region.
95  */
guest_code(void)96 static void guest_code(void)
97 {
98 	uint64_t addr;
99 	int i;
100 
101 	/*
102 	 * On s390x, all pages of a 1M segment are initially marked as dirty
103 	 * when a page of the segment is written to for the very first time.
104 	 * To compensate this specialty in this test, we need to touch all
105 	 * pages during the first iteration.
106 	 */
107 	for (i = 0; i < guest_num_pages; i++) {
108 		addr = guest_test_virt_mem + i * guest_page_size;
109 		vcpu_arch_put_guest(*(uint64_t *)addr, READ_ONCE(iteration));
110 	}
111 
112 	while (true) {
113 		for (i = 0; i < TEST_PAGES_PER_LOOP; i++) {
114 			addr = guest_test_virt_mem;
115 			addr += (guest_random_u64(&guest_rng) % guest_num_pages)
116 				* guest_page_size;
117 			addr = align_down(addr, host_page_size);
118 
119 			vcpu_arch_put_guest(*(uint64_t *)addr, READ_ONCE(iteration));
120 		}
121 
122 		GUEST_SYNC(1);
123 	}
124 }
125 
126 /* Host variables */
127 static bool host_quit;
128 
129 /* Points to the test VM memory region on which we track dirty logs */
130 static void *host_test_mem;
131 static uint64_t host_num_pages;
132 
133 /* For statistics only */
134 static uint64_t host_dirty_count;
135 static uint64_t host_clear_count;
136 static uint64_t host_track_next_count;
137 
138 /* Whether dirty ring reset is requested, or finished */
139 static sem_t sem_vcpu_stop;
140 static sem_t sem_vcpu_cont;
141 /*
142  * This is only set by main thread, and only cleared by vcpu thread.  It is
143  * used to request vcpu thread to stop at the next GUEST_SYNC, since GUEST_SYNC
144  * is the only place that we'll guarantee both "dirty bit" and "dirty data"
145  * will match.  E.g., SIG_IPI won't guarantee that if the vcpu is interrupted
146  * after setting dirty bit but before the data is written.
147  */
148 static atomic_t vcpu_sync_stop_requested;
149 /*
150  * This is updated by the vcpu thread to tell the host whether it's a
151  * ring-full event.  It should only be read until a sem_wait() of
152  * sem_vcpu_stop and before vcpu continues to run.
153  */
154 static bool dirty_ring_vcpu_ring_full;
155 /*
156  * This is only used for verifying the dirty pages.  Dirty ring has a very
157  * tricky case when the ring just got full, kvm will do userspace exit due to
158  * ring full.  When that happens, the very last PFN is set but actually the
159  * data is not changed (the guest WRITE is not really applied yet), because
160  * we found that the dirty ring is full, refused to continue the vcpu, and
161  * recorded the dirty gfn with the old contents.
162  *
163  * For this specific case, it's safe to skip checking this pfn for this
164  * bit, because it's a redundant bit, and when the write happens later the bit
165  * will be set again.  We use this variable to always keep track of the latest
166  * dirty gfn we've collected, so that if a mismatch of data found later in the
167  * verifying process, we let it pass.
168  */
169 static uint64_t dirty_ring_last_page;
170 
171 enum log_mode_t {
172 	/* Only use KVM_GET_DIRTY_LOG for logging */
173 	LOG_MODE_DIRTY_LOG = 0,
174 
175 	/* Use both KVM_[GET|CLEAR]_DIRTY_LOG for logging */
176 	LOG_MODE_CLEAR_LOG = 1,
177 
178 	/* Use dirty ring for logging */
179 	LOG_MODE_DIRTY_RING = 2,
180 
181 	LOG_MODE_NUM,
182 
183 	/* Run all supported modes */
184 	LOG_MODE_ALL = LOG_MODE_NUM,
185 };
186 
187 /* Mode of logging to test.  Default is to run all supported modes */
188 static enum log_mode_t host_log_mode_option = LOG_MODE_ALL;
189 /* Logging mode for current run */
190 static enum log_mode_t host_log_mode;
191 static pthread_t vcpu_thread;
192 static uint32_t test_dirty_ring_count = TEST_DIRTY_RING_COUNT;
193 
vcpu_kick(void)194 static void vcpu_kick(void)
195 {
196 	pthread_kill(vcpu_thread, SIG_IPI);
197 }
198 
199 /*
200  * In our test we do signal tricks, let's use a better version of
201  * sem_wait to avoid signal interrupts
202  */
sem_wait_until(sem_t * sem)203 static void sem_wait_until(sem_t *sem)
204 {
205 	int ret;
206 
207 	do
208 		ret = sem_wait(sem);
209 	while (ret == -1 && errno == EINTR);
210 }
211 
clear_log_supported(void)212 static bool clear_log_supported(void)
213 {
214 	return kvm_has_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
215 }
216 
clear_log_create_vm_done(struct kvm_vm * vm)217 static void clear_log_create_vm_done(struct kvm_vm *vm)
218 {
219 	u64 manual_caps;
220 
221 	manual_caps = kvm_check_cap(KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2);
222 	TEST_ASSERT(manual_caps, "MANUAL_CAPS is zero!");
223 	manual_caps &= (KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE |
224 			KVM_DIRTY_LOG_INITIALLY_SET);
225 	vm_enable_cap(vm, KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2, manual_caps);
226 }
227 
dirty_log_collect_dirty_pages(struct kvm_vcpu * vcpu,int slot,void * bitmap,uint32_t num_pages,uint32_t * unused)228 static void dirty_log_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
229 					  void *bitmap, uint32_t num_pages,
230 					  uint32_t *unused)
231 {
232 	kvm_vm_get_dirty_log(vcpu->vm, slot, bitmap);
233 }
234 
clear_log_collect_dirty_pages(struct kvm_vcpu * vcpu,int slot,void * bitmap,uint32_t num_pages,uint32_t * unused)235 static void clear_log_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
236 					  void *bitmap, uint32_t num_pages,
237 					  uint32_t *unused)
238 {
239 	kvm_vm_get_dirty_log(vcpu->vm, slot, bitmap);
240 	kvm_vm_clear_dirty_log(vcpu->vm, slot, bitmap, 0, num_pages);
241 }
242 
243 /* Should only be called after a GUEST_SYNC */
vcpu_handle_sync_stop(void)244 static void vcpu_handle_sync_stop(void)
245 {
246 	if (atomic_read(&vcpu_sync_stop_requested)) {
247 		/* It means main thread is sleeping waiting */
248 		atomic_set(&vcpu_sync_stop_requested, false);
249 		sem_post(&sem_vcpu_stop);
250 		sem_wait_until(&sem_vcpu_cont);
251 	}
252 }
253 
default_after_vcpu_run(struct kvm_vcpu * vcpu,int ret,int err)254 static void default_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
255 {
256 	struct kvm_run *run = vcpu->run;
257 
258 	TEST_ASSERT(ret == 0 || (ret == -1 && err == EINTR),
259 		    "vcpu run failed: errno=%d", err);
260 
261 	TEST_ASSERT(get_ucall(vcpu, NULL) == UCALL_SYNC,
262 		    "Invalid guest sync status: exit_reason=%s",
263 		    exit_reason_str(run->exit_reason));
264 
265 	vcpu_handle_sync_stop();
266 }
267 
dirty_ring_supported(void)268 static bool dirty_ring_supported(void)
269 {
270 	return (kvm_has_cap(KVM_CAP_DIRTY_LOG_RING) ||
271 		kvm_has_cap(KVM_CAP_DIRTY_LOG_RING_ACQ_REL));
272 }
273 
dirty_ring_create_vm_done(struct kvm_vm * vm)274 static void dirty_ring_create_vm_done(struct kvm_vm *vm)
275 {
276 	uint64_t pages;
277 	uint32_t limit;
278 
279 	/*
280 	 * We rely on vcpu exit due to full dirty ring state. Adjust
281 	 * the ring buffer size to ensure we're able to reach the
282 	 * full dirty ring state.
283 	 */
284 	pages = (1ul << (DIRTY_MEM_BITS - vm->page_shift)) + 3;
285 	pages = vm_adjust_num_guest_pages(vm->mode, pages);
286 	if (vm->page_size < getpagesize())
287 		pages = vm_num_host_pages(vm->mode, pages);
288 
289 	limit = 1 << (31 - __builtin_clz(pages));
290 	test_dirty_ring_count = 1 << (31 - __builtin_clz(test_dirty_ring_count));
291 	test_dirty_ring_count = min(limit, test_dirty_ring_count);
292 	pr_info("dirty ring count: 0x%x\n", test_dirty_ring_count);
293 
294 	/*
295 	 * Switch to dirty ring mode after VM creation but before any
296 	 * of the vcpu creation.
297 	 */
298 	vm_enable_dirty_ring(vm, test_dirty_ring_count *
299 			     sizeof(struct kvm_dirty_gfn));
300 }
301 
dirty_gfn_is_dirtied(struct kvm_dirty_gfn * gfn)302 static inline bool dirty_gfn_is_dirtied(struct kvm_dirty_gfn *gfn)
303 {
304 	return smp_load_acquire(&gfn->flags) == KVM_DIRTY_GFN_F_DIRTY;
305 }
306 
dirty_gfn_set_collected(struct kvm_dirty_gfn * gfn)307 static inline void dirty_gfn_set_collected(struct kvm_dirty_gfn *gfn)
308 {
309 	smp_store_release(&gfn->flags, KVM_DIRTY_GFN_F_RESET);
310 }
311 
dirty_ring_collect_one(struct kvm_dirty_gfn * dirty_gfns,int slot,void * bitmap,uint32_t num_pages,uint32_t * fetch_index)312 static uint32_t dirty_ring_collect_one(struct kvm_dirty_gfn *dirty_gfns,
313 				       int slot, void *bitmap,
314 				       uint32_t num_pages, uint32_t *fetch_index)
315 {
316 	struct kvm_dirty_gfn *cur;
317 	uint32_t count = 0;
318 
319 	while (true) {
320 		cur = &dirty_gfns[*fetch_index % test_dirty_ring_count];
321 		if (!dirty_gfn_is_dirtied(cur))
322 			break;
323 		TEST_ASSERT(cur->slot == slot, "Slot number didn't match: "
324 			    "%u != %u", cur->slot, slot);
325 		TEST_ASSERT(cur->offset < num_pages, "Offset overflow: "
326 			    "0x%llx >= 0x%x", cur->offset, num_pages);
327 		//pr_info("fetch 0x%x page %llu\n", *fetch_index, cur->offset);
328 		__set_bit_le(cur->offset, bitmap);
329 		dirty_ring_last_page = cur->offset;
330 		dirty_gfn_set_collected(cur);
331 		(*fetch_index)++;
332 		count++;
333 	}
334 
335 	return count;
336 }
337 
dirty_ring_wait_vcpu(void)338 static void dirty_ring_wait_vcpu(void)
339 {
340 	/* This makes sure that hardware PML cache flushed */
341 	vcpu_kick();
342 	sem_wait_until(&sem_vcpu_stop);
343 }
344 
dirty_ring_continue_vcpu(void)345 static void dirty_ring_continue_vcpu(void)
346 {
347 	pr_info("Notifying vcpu to continue\n");
348 	sem_post(&sem_vcpu_cont);
349 }
350 
dirty_ring_collect_dirty_pages(struct kvm_vcpu * vcpu,int slot,void * bitmap,uint32_t num_pages,uint32_t * ring_buf_idx)351 static void dirty_ring_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
352 					   void *bitmap, uint32_t num_pages,
353 					   uint32_t *ring_buf_idx)
354 {
355 	uint32_t count = 0, cleared;
356 	bool continued_vcpu = false;
357 
358 	dirty_ring_wait_vcpu();
359 
360 	if (!dirty_ring_vcpu_ring_full) {
361 		/*
362 		 * This is not a ring-full event, it's safe to allow
363 		 * vcpu to continue
364 		 */
365 		dirty_ring_continue_vcpu();
366 		continued_vcpu = true;
367 	}
368 
369 	/* Only have one vcpu */
370 	count = dirty_ring_collect_one(vcpu_map_dirty_ring(vcpu),
371 				       slot, bitmap, num_pages,
372 				       ring_buf_idx);
373 
374 	cleared = kvm_vm_reset_dirty_ring(vcpu->vm);
375 
376 	/*
377 	 * Cleared pages should be the same as collected, as KVM is supposed to
378 	 * clear only the entries that have been harvested.
379 	 */
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 
dirty_ring_after_vcpu_run(struct kvm_vcpu * vcpu,int ret,int err)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",
414 			    exit_reason_str(run->exit_reason));
415 	}
416 }
417 
418 struct log_mode {
419 	const char *name;
420 	/* Return true if this mode is supported, otherwise false */
421 	bool (*supported)(void);
422 	/* Hook when the vm creation is done (before vcpu creation) */
423 	void (*create_vm_done)(struct kvm_vm *vm);
424 	/* Hook to collect the dirty pages into the bitmap provided */
425 	void (*collect_dirty_pages) (struct kvm_vcpu *vcpu, int slot,
426 				     void *bitmap, uint32_t num_pages,
427 				     uint32_t *ring_buf_idx);
428 	/* Hook to call when after each vcpu run */
429 	void (*after_vcpu_run)(struct kvm_vcpu *vcpu, int ret, int err);
430 } log_modes[LOG_MODE_NUM] = {
431 	{
432 		.name = "dirty-log",
433 		.collect_dirty_pages = dirty_log_collect_dirty_pages,
434 		.after_vcpu_run = default_after_vcpu_run,
435 	},
436 	{
437 		.name = "clear-log",
438 		.supported = clear_log_supported,
439 		.create_vm_done = clear_log_create_vm_done,
440 		.collect_dirty_pages = clear_log_collect_dirty_pages,
441 		.after_vcpu_run = default_after_vcpu_run,
442 	},
443 	{
444 		.name = "dirty-ring",
445 		.supported = dirty_ring_supported,
446 		.create_vm_done = dirty_ring_create_vm_done,
447 		.collect_dirty_pages = dirty_ring_collect_dirty_pages,
448 		.after_vcpu_run = dirty_ring_after_vcpu_run,
449 	},
450 };
451 
452 /*
453  * We use this bitmap to track some pages that should have its dirty
454  * bit set in the _next_ iteration.  For example, if we detected the
455  * page value changed to current iteration but at the same time the
456  * page bit is cleared in the latest bitmap, then the system must
457  * report that write in the next get dirty log call.
458  */
459 static unsigned long *host_bmap_track;
460 
log_modes_dump(void)461 static void log_modes_dump(void)
462 {
463 	int i;
464 
465 	printf("all");
466 	for (i = 0; i < LOG_MODE_NUM; i++)
467 		printf(", %s", log_modes[i].name);
468 	printf("\n");
469 }
470 
log_mode_supported(void)471 static bool log_mode_supported(void)
472 {
473 	struct log_mode *mode = &log_modes[host_log_mode];
474 
475 	if (mode->supported)
476 		return mode->supported();
477 
478 	return true;
479 }
480 
log_mode_create_vm_done(struct kvm_vm * vm)481 static void log_mode_create_vm_done(struct kvm_vm *vm)
482 {
483 	struct log_mode *mode = &log_modes[host_log_mode];
484 
485 	if (mode->create_vm_done)
486 		mode->create_vm_done(vm);
487 }
488 
log_mode_collect_dirty_pages(struct kvm_vcpu * vcpu,int slot,void * bitmap,uint32_t num_pages,uint32_t * ring_buf_idx)489 static void log_mode_collect_dirty_pages(struct kvm_vcpu *vcpu, int slot,
490 					 void *bitmap, uint32_t num_pages,
491 					 uint32_t *ring_buf_idx)
492 {
493 	struct log_mode *mode = &log_modes[host_log_mode];
494 
495 	TEST_ASSERT(mode->collect_dirty_pages != NULL,
496 		    "collect_dirty_pages() is required for any log mode!");
497 	mode->collect_dirty_pages(vcpu, slot, bitmap, num_pages, ring_buf_idx);
498 }
499 
log_mode_after_vcpu_run(struct kvm_vcpu * vcpu,int ret,int err)500 static void log_mode_after_vcpu_run(struct kvm_vcpu *vcpu, int ret, int err)
501 {
502 	struct log_mode *mode = &log_modes[host_log_mode];
503 
504 	if (mode->after_vcpu_run)
505 		mode->after_vcpu_run(vcpu, ret, err);
506 }
507 
vcpu_worker(void * data)508 static void *vcpu_worker(void *data)
509 {
510 	int ret;
511 	struct kvm_vcpu *vcpu = data;
512 	uint64_t pages_count = 0;
513 	struct kvm_signal_mask *sigmask = alloca(offsetof(struct kvm_signal_mask, sigset)
514 						 + sizeof(sigset_t));
515 	sigset_t *sigset = (sigset_t *) &sigmask->sigset;
516 
517 	/*
518 	 * SIG_IPI is unblocked atomically while in KVM_RUN.  It causes the
519 	 * ioctl to return with -EINTR, but it is still pending and we need
520 	 * to accept it with the sigwait.
521 	 */
522 	sigmask->len = 8;
523 	pthread_sigmask(0, NULL, sigset);
524 	sigdelset(sigset, SIG_IPI);
525 	vcpu_ioctl(vcpu, KVM_SET_SIGNAL_MASK, sigmask);
526 
527 	sigemptyset(sigset);
528 	sigaddset(sigset, SIG_IPI);
529 
530 	while (!READ_ONCE(host_quit)) {
531 		/* Clear any existing kick signals */
532 		pages_count += TEST_PAGES_PER_LOOP;
533 		/* Let the guest dirty the random pages */
534 		ret = __vcpu_run(vcpu);
535 		if (ret == -1 && errno == EINTR) {
536 			int sig = -1;
537 			sigwait(sigset, &sig);
538 			assert(sig == SIG_IPI);
539 		}
540 		log_mode_after_vcpu_run(vcpu, ret, errno);
541 	}
542 
543 	pr_info("Dirtied %"PRIu64" pages\n", pages_count);
544 
545 	return NULL;
546 }
547 
vm_dirty_log_verify(enum vm_guest_mode mode,unsigned long * bmap)548 static void vm_dirty_log_verify(enum vm_guest_mode mode, unsigned long *bmap)
549 {
550 	uint64_t step = vm_num_host_pages(mode, 1);
551 	uint64_t page;
552 	uint64_t *value_ptr;
553 	uint64_t min_iter = 0;
554 
555 	for (page = 0; page < host_num_pages; page += step) {
556 		value_ptr = host_test_mem + page * host_page_size;
557 
558 		/* If this is a special page that we were tracking... */
559 		if (__test_and_clear_bit_le(page, host_bmap_track)) {
560 			host_track_next_count++;
561 			TEST_ASSERT(test_bit_le(page, bmap),
562 				    "Page %"PRIu64" should have its dirty bit "
563 				    "set in this iteration but it is missing",
564 				    page);
565 		}
566 
567 		if (__test_and_clear_bit_le(page, bmap)) {
568 			bool matched;
569 
570 			host_dirty_count++;
571 
572 			/*
573 			 * If the bit is set, the value written onto
574 			 * the corresponding page should be either the
575 			 * previous iteration number or the current one.
576 			 */
577 			matched = (*value_ptr == iteration ||
578 				   *value_ptr == iteration - 1);
579 
580 			if (host_log_mode == LOG_MODE_DIRTY_RING && !matched) {
581 				if (*value_ptr == iteration - 2 && min_iter <= iteration - 2) {
582 					/*
583 					 * Short answer: this case is special
584 					 * only for dirty ring test where the
585 					 * page is the last page before a kvm
586 					 * dirty ring full in iteration N-2.
587 					 *
588 					 * Long answer: Assuming ring size R,
589 					 * one possible condition is:
590 					 *
591 					 *      main thr       vcpu thr
592 					 *      --------       --------
593 					 *    iter=1
594 					 *                   write 1 to page 0~(R-1)
595 					 *                   full, vmexit
596 					 *    collect 0~(R-1)
597 					 *    kick vcpu
598 					 *                   write 1 to (R-1)~(2R-2)
599 					 *                   full, vmexit
600 					 *    iter=2
601 					 *    collect (R-1)~(2R-2)
602 					 *    kick vcpu
603 					 *                   write 1 to (2R-2)
604 					 *                   (NOTE!!! "1" cached in cpu reg)
605 					 *                   write 2 to (2R-1)~(3R-3)
606 					 *                   full, vmexit
607 					 *    iter=3
608 					 *    collect (2R-2)~(3R-3)
609 					 *    (here if we read value on page
610 					 *     "2R-2" is 1, while iter=3!!!)
611 					 *
612 					 * This however can only happen once per iteration.
613 					 */
614 					min_iter = iteration - 1;
615 					continue;
616 				} else if (page == dirty_ring_last_page) {
617 					/*
618 					 * Please refer to comments in
619 					 * dirty_ring_last_page.
620 					 */
621 					continue;
622 				}
623 			}
624 
625 			TEST_ASSERT(matched,
626 				    "Set page %"PRIu64" value %"PRIu64
627 				    " incorrect (iteration=%"PRIu64")",
628 				    page, *value_ptr, iteration);
629 		} else {
630 			host_clear_count++;
631 			/*
632 			 * If cleared, the value written can be any
633 			 * value smaller or equals to the iteration
634 			 * number.  Note that the value can be exactly
635 			 * (iteration-1) if that write can happen
636 			 * like this:
637 			 *
638 			 * (1) increase loop count to "iteration-1"
639 			 * (2) write to page P happens (with value
640 			 *     "iteration-1")
641 			 * (3) get dirty log for "iteration-1"; we'll
642 			 *     see that page P bit is set (dirtied),
643 			 *     and not set the bit in host_bmap_track
644 			 * (4) increase loop count to "iteration"
645 			 *     (which is current iteration)
646 			 * (5) get dirty log for current iteration,
647 			 *     we'll see that page P is cleared, with
648 			 *     value "iteration-1".
649 			 */
650 			TEST_ASSERT(*value_ptr <= iteration,
651 				    "Clear page %"PRIu64" value %"PRIu64
652 				    " incorrect (iteration=%"PRIu64")",
653 				    page, *value_ptr, iteration);
654 			if (*value_ptr == iteration) {
655 				/*
656 				 * This page is _just_ modified; it
657 				 * should report its dirtyness in the
658 				 * next run
659 				 */
660 				__set_bit_le(page, host_bmap_track);
661 			}
662 		}
663 	}
664 }
665 
create_vm(enum vm_guest_mode mode,struct kvm_vcpu ** vcpu,uint64_t extra_mem_pages,void * guest_code)666 static struct kvm_vm *create_vm(enum vm_guest_mode mode, struct kvm_vcpu **vcpu,
667 				uint64_t extra_mem_pages, void *guest_code)
668 {
669 	struct kvm_vm *vm;
670 
671 	pr_info("Testing guest mode: %s\n", vm_guest_mode_string(mode));
672 
673 	vm = __vm_create(VM_SHAPE(mode), 1, extra_mem_pages);
674 
675 	log_mode_create_vm_done(vm);
676 	*vcpu = vm_vcpu_add(vm, 0, guest_code);
677 	return vm;
678 }
679 
680 struct test_params {
681 	unsigned long iterations;
682 	unsigned long interval;
683 	uint64_t phys_offset;
684 };
685 
run_test(enum vm_guest_mode mode,void * arg)686 static void run_test(enum vm_guest_mode mode, void *arg)
687 {
688 	struct test_params *p = arg;
689 	struct kvm_vcpu *vcpu;
690 	struct kvm_vm *vm;
691 	unsigned long *bmap;
692 	uint32_t ring_buf_idx = 0;
693 	int sem_val;
694 
695 	if (!log_mode_supported()) {
696 		print_skip("Log mode '%s' not supported",
697 			   log_modes[host_log_mode].name);
698 		return;
699 	}
700 
701 	/*
702 	 * We reserve page table for 2 times of extra dirty mem which
703 	 * will definitely cover the original (1G+) test range.  Here
704 	 * we do the calculation with 4K page size which is the
705 	 * smallest so the page number will be enough for all archs
706 	 * (e.g., 64K page size guest will need even less memory for
707 	 * page tables).
708 	 */
709 	vm = create_vm(mode, &vcpu,
710 		       2ul << (DIRTY_MEM_BITS - PAGE_SHIFT_4K), guest_code);
711 
712 	guest_page_size = vm->page_size;
713 	/*
714 	 * A little more than 1G of guest page sized pages.  Cover the
715 	 * case where the size is not aligned to 64 pages.
716 	 */
717 	guest_num_pages = (1ul << (DIRTY_MEM_BITS - vm->page_shift)) + 3;
718 	guest_num_pages = vm_adjust_num_guest_pages(mode, guest_num_pages);
719 
720 	host_page_size = getpagesize();
721 	host_num_pages = vm_num_host_pages(mode, guest_num_pages);
722 
723 	if (!p->phys_offset) {
724 		guest_test_phys_mem = (vm->max_gfn - guest_num_pages) *
725 				      guest_page_size;
726 		guest_test_phys_mem = align_down(guest_test_phys_mem, host_page_size);
727 	} else {
728 		guest_test_phys_mem = p->phys_offset;
729 	}
730 
731 #ifdef __s390x__
732 	/* Align to 1M (segment size) */
733 	guest_test_phys_mem = align_down(guest_test_phys_mem, 1 << 20);
734 #endif
735 
736 	pr_info("guest physical test memory offset: 0x%lx\n", guest_test_phys_mem);
737 
738 	bmap = bitmap_zalloc(host_num_pages);
739 	host_bmap_track = bitmap_zalloc(host_num_pages);
740 
741 	/* Add an extra memory slot for testing dirty logging */
742 	vm_userspace_mem_region_add(vm, VM_MEM_SRC_ANONYMOUS,
743 				    guest_test_phys_mem,
744 				    TEST_MEM_SLOT_INDEX,
745 				    guest_num_pages,
746 				    KVM_MEM_LOG_DIRTY_PAGES);
747 
748 	/* Do mapping for the dirty track memory slot */
749 	virt_map(vm, guest_test_virt_mem, guest_test_phys_mem, guest_num_pages);
750 
751 	/* Cache the HVA pointer of the region */
752 	host_test_mem = addr_gpa2hva(vm, (vm_paddr_t)guest_test_phys_mem);
753 
754 	/* Export the shared variables to the guest */
755 	sync_global_to_guest(vm, host_page_size);
756 	sync_global_to_guest(vm, guest_page_size);
757 	sync_global_to_guest(vm, guest_test_virt_mem);
758 	sync_global_to_guest(vm, guest_num_pages);
759 
760 	/* Start the iterations */
761 	iteration = 1;
762 	sync_global_to_guest(vm, iteration);
763 	WRITE_ONCE(host_quit, false);
764 	host_dirty_count = 0;
765 	host_clear_count = 0;
766 	host_track_next_count = 0;
767 	WRITE_ONCE(dirty_ring_vcpu_ring_full, false);
768 
769 	/*
770 	 * Ensure the previous iteration didn't leave a dangling semaphore, i.e.
771 	 * that the main task and vCPU worker were synchronized and completed
772 	 * verification of all iterations.
773 	 */
774 	sem_getvalue(&sem_vcpu_stop, &sem_val);
775 	TEST_ASSERT_EQ(sem_val, 0);
776 	sem_getvalue(&sem_vcpu_cont, &sem_val);
777 	TEST_ASSERT_EQ(sem_val, 0);
778 
779 	pthread_create(&vcpu_thread, NULL, vcpu_worker, vcpu);
780 
781 	while (iteration < p->iterations) {
782 		/* Give the vcpu thread some time to dirty some pages */
783 		usleep(p->interval * 1000);
784 		log_mode_collect_dirty_pages(vcpu, TEST_MEM_SLOT_INDEX,
785 					     bmap, host_num_pages,
786 					     &ring_buf_idx);
787 
788 		/*
789 		 * See vcpu_sync_stop_requested definition for details on why
790 		 * we need to stop vcpu when verify data.
791 		 */
792 		atomic_set(&vcpu_sync_stop_requested, true);
793 		sem_wait_until(&sem_vcpu_stop);
794 		/*
795 		 * NOTE: for dirty ring, it's possible that we didn't stop at
796 		 * GUEST_SYNC but instead we stopped because ring is full;
797 		 * that's okay too because ring full means we're only missing
798 		 * the flush of the last page, and since we handle the last
799 		 * page specially verification will succeed anyway.
800 		 */
801 		assert(host_log_mode == LOG_MODE_DIRTY_RING ||
802 		       atomic_read(&vcpu_sync_stop_requested) == false);
803 		vm_dirty_log_verify(mode, bmap);
804 
805 		/*
806 		 * Set host_quit before sem_vcpu_cont in the final iteration to
807 		 * ensure that the vCPU worker doesn't resume the guest.  As
808 		 * above, the dirty ring test may stop and wait even when not
809 		 * explicitly request to do so, i.e. would hang waiting for a
810 		 * "continue" if it's allowed to resume the guest.
811 		 */
812 		if (++iteration == p->iterations)
813 			WRITE_ONCE(host_quit, true);
814 
815 		sem_post(&sem_vcpu_cont);
816 		sync_global_to_guest(vm, iteration);
817 	}
818 
819 	pthread_join(vcpu_thread, NULL);
820 
821 	pr_info("Total bits checked: dirty (%"PRIu64"), clear (%"PRIu64"), "
822 		"track_next (%"PRIu64")\n", host_dirty_count, host_clear_count,
823 		host_track_next_count);
824 
825 	free(bmap);
826 	free(host_bmap_track);
827 	kvm_vm_free(vm);
828 }
829 
help(char * name)830 static void help(char *name)
831 {
832 	puts("");
833 	printf("usage: %s [-h] [-i iterations] [-I interval] "
834 	       "[-p offset] [-m mode]\n", name);
835 	puts("");
836 	printf(" -c: hint to dirty ring size, in number of entries\n");
837 	printf("     (only useful for dirty-ring test; default: %"PRIu32")\n",
838 	       TEST_DIRTY_RING_COUNT);
839 	printf(" -i: specify iteration counts (default: %"PRIu64")\n",
840 	       TEST_HOST_LOOP_N);
841 	printf(" -I: specify interval in ms (default: %"PRIu64" ms)\n",
842 	       TEST_HOST_LOOP_INTERVAL);
843 	printf(" -p: specify guest physical test memory offset\n"
844 	       "     Warning: a low offset can conflict with the loaded test code.\n");
845 	printf(" -M: specify the host logging mode "
846 	       "(default: run all log modes).  Supported modes: \n\t");
847 	log_modes_dump();
848 	guest_modes_help();
849 	puts("");
850 	exit(0);
851 }
852 
main(int argc,char * argv[])853 int main(int argc, char *argv[])
854 {
855 	struct test_params p = {
856 		.iterations = TEST_HOST_LOOP_N,
857 		.interval = TEST_HOST_LOOP_INTERVAL,
858 	};
859 	int opt, i;
860 	sigset_t sigset;
861 
862 	sem_init(&sem_vcpu_stop, 0, 0);
863 	sem_init(&sem_vcpu_cont, 0, 0);
864 
865 	guest_modes_append_default();
866 
867 	while ((opt = getopt(argc, argv, "c:hi:I:p:m:M:")) != -1) {
868 		switch (opt) {
869 		case 'c':
870 			test_dirty_ring_count = strtol(optarg, NULL, 10);
871 			break;
872 		case 'i':
873 			p.iterations = strtol(optarg, NULL, 10);
874 			break;
875 		case 'I':
876 			p.interval = strtol(optarg, NULL, 10);
877 			break;
878 		case 'p':
879 			p.phys_offset = strtoull(optarg, NULL, 0);
880 			break;
881 		case 'm':
882 			guest_modes_cmdline(optarg);
883 			break;
884 		case 'M':
885 			if (!strcmp(optarg, "all")) {
886 				host_log_mode_option = LOG_MODE_ALL;
887 				break;
888 			}
889 			for (i = 0; i < LOG_MODE_NUM; i++) {
890 				if (!strcmp(optarg, log_modes[i].name)) {
891 					pr_info("Setting log mode to: '%s'\n",
892 						optarg);
893 					host_log_mode_option = i;
894 					break;
895 				}
896 			}
897 			if (i == LOG_MODE_NUM) {
898 				printf("Log mode '%s' invalid. Please choose "
899 				       "from: ", optarg);
900 				log_modes_dump();
901 				exit(1);
902 			}
903 			break;
904 		case 'h':
905 		default:
906 			help(argv[0]);
907 			break;
908 		}
909 	}
910 
911 	TEST_ASSERT(p.iterations > 2, "Iterations must be greater than two");
912 	TEST_ASSERT(p.interval > 0, "Interval must be greater than zero");
913 
914 	pr_info("Test iterations: %"PRIu64", interval: %"PRIu64" (ms)\n",
915 		p.iterations, p.interval);
916 
917 	srandom(time(0));
918 
919 	/* Ensure that vCPU threads start with SIG_IPI blocked.  */
920 	sigemptyset(&sigset);
921 	sigaddset(&sigset, SIG_IPI);
922 	pthread_sigmask(SIG_BLOCK, &sigset, NULL);
923 
924 	if (host_log_mode_option == LOG_MODE_ALL) {
925 		/* Run each log mode */
926 		for (i = 0; i < LOG_MODE_NUM; i++) {
927 			pr_info("Testing Log Mode '%s'\n", log_modes[i].name);
928 			host_log_mode = i;
929 			for_each_guest_mode(run_test, &p);
930 		}
931 	} else {
932 		host_log_mode = host_log_mode_option;
933 		for_each_guest_mode(run_test, &p);
934 	}
935 
936 	return 0;
937 }
938