coroutine: trim down nesting level in perf_nesting test
[qemu.git] / dump.c
bloba25f5096eb51b63f7bd068874e5e20f1cd24beb3
1 /*
2 * QEMU dump
4 * Copyright Fujitsu, Corp. 2011, 2012
6 * Authors:
7 * Wen Congyang <wency@cn.fujitsu.com>
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
14 #include "qemu-common.h"
15 #include "elf.h"
16 #include "cpu.h"
17 #include "exec/cpu-all.h"
18 #include "exec/hwaddr.h"
19 #include "monitor/monitor.h"
20 #include "sysemu/kvm.h"
21 #include "sysemu/dump.h"
22 #include "sysemu/sysemu.h"
23 #include "sysemu/memory_mapping.h"
24 #include "qapi/error.h"
25 #include "qmp-commands.h"
26 #include "exec/gdbstub.h"
28 static uint16_t cpu_convert_to_target16(uint16_t val, int endian)
30 if (endian == ELFDATA2LSB) {
31 val = cpu_to_le16(val);
32 } else {
33 val = cpu_to_be16(val);
36 return val;
39 static uint32_t cpu_convert_to_target32(uint32_t val, int endian)
41 if (endian == ELFDATA2LSB) {
42 val = cpu_to_le32(val);
43 } else {
44 val = cpu_to_be32(val);
47 return val;
50 static uint64_t cpu_convert_to_target64(uint64_t val, int endian)
52 if (endian == ELFDATA2LSB) {
53 val = cpu_to_le64(val);
54 } else {
55 val = cpu_to_be64(val);
58 return val;
61 typedef struct DumpState {
62 ArchDumpInfo dump_info;
63 MemoryMappingList list;
64 uint16_t phdr_num;
65 uint32_t sh_info;
66 bool have_section;
67 bool resume;
68 size_t note_size;
69 hwaddr memory_offset;
70 int fd;
72 RAMBlock *block;
73 ram_addr_t start;
74 bool has_filter;
75 int64_t begin;
76 int64_t length;
77 Error **errp;
78 } DumpState;
80 static int dump_cleanup(DumpState *s)
82 int ret = 0;
84 memory_mapping_list_free(&s->list);
85 if (s->fd != -1) {
86 close(s->fd);
88 if (s->resume) {
89 vm_start();
92 return ret;
95 static void dump_error(DumpState *s, const char *reason)
97 dump_cleanup(s);
100 static int fd_write_vmcore(void *buf, size_t size, void *opaque)
102 DumpState *s = opaque;
103 size_t written_size;
105 written_size = qemu_write_full(s->fd, buf, size);
106 if (written_size != size) {
107 return -1;
110 return 0;
113 static int write_elf64_header(DumpState *s)
115 Elf64_Ehdr elf_header;
116 int ret;
117 int endian = s->dump_info.d_endian;
119 memset(&elf_header, 0, sizeof(Elf64_Ehdr));
120 memcpy(&elf_header, ELFMAG, SELFMAG);
121 elf_header.e_ident[EI_CLASS] = ELFCLASS64;
122 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
123 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
124 elf_header.e_type = cpu_convert_to_target16(ET_CORE, endian);
125 elf_header.e_machine = cpu_convert_to_target16(s->dump_info.d_machine,
126 endian);
127 elf_header.e_version = cpu_convert_to_target32(EV_CURRENT, endian);
128 elf_header.e_ehsize = cpu_convert_to_target16(sizeof(elf_header), endian);
129 elf_header.e_phoff = cpu_convert_to_target64(sizeof(Elf64_Ehdr), endian);
130 elf_header.e_phentsize = cpu_convert_to_target16(sizeof(Elf64_Phdr),
131 endian);
132 elf_header.e_phnum = cpu_convert_to_target16(s->phdr_num, endian);
133 if (s->have_section) {
134 uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
136 elf_header.e_shoff = cpu_convert_to_target64(shoff, endian);
137 elf_header.e_shentsize = cpu_convert_to_target16(sizeof(Elf64_Shdr),
138 endian);
139 elf_header.e_shnum = cpu_convert_to_target16(1, endian);
142 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
143 if (ret < 0) {
144 dump_error(s, "dump: failed to write elf header.\n");
145 return -1;
148 return 0;
151 static int write_elf32_header(DumpState *s)
153 Elf32_Ehdr elf_header;
154 int ret;
155 int endian = s->dump_info.d_endian;
157 memset(&elf_header, 0, sizeof(Elf32_Ehdr));
158 memcpy(&elf_header, ELFMAG, SELFMAG);
159 elf_header.e_ident[EI_CLASS] = ELFCLASS32;
160 elf_header.e_ident[EI_DATA] = endian;
161 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
162 elf_header.e_type = cpu_convert_to_target16(ET_CORE, endian);
163 elf_header.e_machine = cpu_convert_to_target16(s->dump_info.d_machine,
164 endian);
165 elf_header.e_version = cpu_convert_to_target32(EV_CURRENT, endian);
166 elf_header.e_ehsize = cpu_convert_to_target16(sizeof(elf_header), endian);
167 elf_header.e_phoff = cpu_convert_to_target32(sizeof(Elf32_Ehdr), endian);
168 elf_header.e_phentsize = cpu_convert_to_target16(sizeof(Elf32_Phdr),
169 endian);
170 elf_header.e_phnum = cpu_convert_to_target16(s->phdr_num, endian);
171 if (s->have_section) {
172 uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
174 elf_header.e_shoff = cpu_convert_to_target32(shoff, endian);
175 elf_header.e_shentsize = cpu_convert_to_target16(sizeof(Elf32_Shdr),
176 endian);
177 elf_header.e_shnum = cpu_convert_to_target16(1, endian);
180 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
181 if (ret < 0) {
182 dump_error(s, "dump: failed to write elf header.\n");
183 return -1;
186 return 0;
189 static int write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
190 int phdr_index, hwaddr offset)
192 Elf64_Phdr phdr;
193 int ret;
194 int endian = s->dump_info.d_endian;
196 memset(&phdr, 0, sizeof(Elf64_Phdr));
197 phdr.p_type = cpu_convert_to_target32(PT_LOAD, endian);
198 phdr.p_offset = cpu_convert_to_target64(offset, endian);
199 phdr.p_paddr = cpu_convert_to_target64(memory_mapping->phys_addr, endian);
200 if (offset == -1) {
201 /* When the memory is not stored into vmcore, offset will be -1 */
202 phdr.p_filesz = 0;
203 } else {
204 phdr.p_filesz = cpu_convert_to_target64(memory_mapping->length, endian);
206 phdr.p_memsz = cpu_convert_to_target64(memory_mapping->length, endian);
207 phdr.p_vaddr = cpu_convert_to_target64(memory_mapping->virt_addr, endian);
209 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
210 if (ret < 0) {
211 dump_error(s, "dump: failed to write program header table.\n");
212 return -1;
215 return 0;
218 static int write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
219 int phdr_index, hwaddr offset)
221 Elf32_Phdr phdr;
222 int ret;
223 int endian = s->dump_info.d_endian;
225 memset(&phdr, 0, sizeof(Elf32_Phdr));
226 phdr.p_type = cpu_convert_to_target32(PT_LOAD, endian);
227 phdr.p_offset = cpu_convert_to_target32(offset, endian);
228 phdr.p_paddr = cpu_convert_to_target32(memory_mapping->phys_addr, endian);
229 if (offset == -1) {
230 /* When the memory is not stored into vmcore, offset will be -1 */
231 phdr.p_filesz = 0;
232 } else {
233 phdr.p_filesz = cpu_convert_to_target32(memory_mapping->length, endian);
235 phdr.p_memsz = cpu_convert_to_target32(memory_mapping->length, endian);
236 phdr.p_vaddr = cpu_convert_to_target32(memory_mapping->virt_addr, endian);
238 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
239 if (ret < 0) {
240 dump_error(s, "dump: failed to write program header table.\n");
241 return -1;
244 return 0;
247 static int write_elf64_note(DumpState *s)
249 Elf64_Phdr phdr;
250 int endian = s->dump_info.d_endian;
251 hwaddr begin = s->memory_offset - s->note_size;
252 int ret;
254 memset(&phdr, 0, sizeof(Elf64_Phdr));
255 phdr.p_type = cpu_convert_to_target32(PT_NOTE, endian);
256 phdr.p_offset = cpu_convert_to_target64(begin, endian);
257 phdr.p_paddr = 0;
258 phdr.p_filesz = cpu_convert_to_target64(s->note_size, endian);
259 phdr.p_memsz = cpu_convert_to_target64(s->note_size, endian);
260 phdr.p_vaddr = 0;
262 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
263 if (ret < 0) {
264 dump_error(s, "dump: failed to write program header table.\n");
265 return -1;
268 return 0;
271 static int write_elf64_notes(DumpState *s)
273 CPUArchState *env;
274 CPUState *cpu;
275 int ret;
276 int id;
278 for (env = first_cpu; env != NULL; env = env->next_cpu) {
279 cpu = ENV_GET_CPU(env);
280 id = cpu_index(cpu);
281 ret = cpu_write_elf64_note(fd_write_vmcore, env, id, s);
282 if (ret < 0) {
283 dump_error(s, "dump: failed to write elf notes.\n");
284 return -1;
288 for (env = first_cpu; env != NULL; env = env->next_cpu) {
289 ret = cpu_write_elf64_qemunote(fd_write_vmcore, env, s);
290 if (ret < 0) {
291 dump_error(s, "dump: failed to write CPU status.\n");
292 return -1;
296 return 0;
299 static int write_elf32_note(DumpState *s)
301 hwaddr begin = s->memory_offset - s->note_size;
302 Elf32_Phdr phdr;
303 int endian = s->dump_info.d_endian;
304 int ret;
306 memset(&phdr, 0, sizeof(Elf32_Phdr));
307 phdr.p_type = cpu_convert_to_target32(PT_NOTE, endian);
308 phdr.p_offset = cpu_convert_to_target32(begin, endian);
309 phdr.p_paddr = 0;
310 phdr.p_filesz = cpu_convert_to_target32(s->note_size, endian);
311 phdr.p_memsz = cpu_convert_to_target32(s->note_size, endian);
312 phdr.p_vaddr = 0;
314 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
315 if (ret < 0) {
316 dump_error(s, "dump: failed to write program header table.\n");
317 return -1;
320 return 0;
323 static int write_elf32_notes(DumpState *s)
325 CPUArchState *env;
326 CPUState *cpu;
327 int ret;
328 int id;
330 for (env = first_cpu; env != NULL; env = env->next_cpu) {
331 cpu = ENV_GET_CPU(env);
332 id = cpu_index(cpu);
333 ret = cpu_write_elf32_note(fd_write_vmcore, env, id, s);
334 if (ret < 0) {
335 dump_error(s, "dump: failed to write elf notes.\n");
336 return -1;
340 for (env = first_cpu; env != NULL; env = env->next_cpu) {
341 ret = cpu_write_elf32_qemunote(fd_write_vmcore, env, s);
342 if (ret < 0) {
343 dump_error(s, "dump: failed to write CPU status.\n");
344 return -1;
348 return 0;
351 static int write_elf_section(DumpState *s, int type)
353 Elf32_Shdr shdr32;
354 Elf64_Shdr shdr64;
355 int endian = s->dump_info.d_endian;
356 int shdr_size;
357 void *shdr;
358 int ret;
360 if (type == 0) {
361 shdr_size = sizeof(Elf32_Shdr);
362 memset(&shdr32, 0, shdr_size);
363 shdr32.sh_info = cpu_convert_to_target32(s->sh_info, endian);
364 shdr = &shdr32;
365 } else {
366 shdr_size = sizeof(Elf64_Shdr);
367 memset(&shdr64, 0, shdr_size);
368 shdr64.sh_info = cpu_convert_to_target32(s->sh_info, endian);
369 shdr = &shdr64;
372 ret = fd_write_vmcore(&shdr, shdr_size, s);
373 if (ret < 0) {
374 dump_error(s, "dump: failed to write section header table.\n");
375 return -1;
378 return 0;
381 static int write_data(DumpState *s, void *buf, int length)
383 int ret;
385 ret = fd_write_vmcore(buf, length, s);
386 if (ret < 0) {
387 dump_error(s, "dump: failed to save memory.\n");
388 return -1;
391 return 0;
394 /* write the memroy to vmcore. 1 page per I/O. */
395 static int write_memory(DumpState *s, RAMBlock *block, ram_addr_t start,
396 int64_t size)
398 int64_t i;
399 int ret;
401 for (i = 0; i < size / TARGET_PAGE_SIZE; i++) {
402 ret = write_data(s, block->host + start + i * TARGET_PAGE_SIZE,
403 TARGET_PAGE_SIZE);
404 if (ret < 0) {
405 return ret;
409 if ((size % TARGET_PAGE_SIZE) != 0) {
410 ret = write_data(s, block->host + start + i * TARGET_PAGE_SIZE,
411 size % TARGET_PAGE_SIZE);
412 if (ret < 0) {
413 return ret;
417 return 0;
420 /* get the memory's offset in the vmcore */
421 static hwaddr get_offset(hwaddr phys_addr,
422 DumpState *s)
424 RAMBlock *block;
425 hwaddr offset = s->memory_offset;
426 int64_t size_in_block, start;
428 if (s->has_filter) {
429 if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
430 return -1;
434 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
435 if (s->has_filter) {
436 if (block->offset >= s->begin + s->length ||
437 block->offset + block->length <= s->begin) {
438 /* This block is out of the range */
439 continue;
442 if (s->begin <= block->offset) {
443 start = block->offset;
444 } else {
445 start = s->begin;
448 size_in_block = block->length - (start - block->offset);
449 if (s->begin + s->length < block->offset + block->length) {
450 size_in_block -= block->offset + block->length -
451 (s->begin + s->length);
453 } else {
454 start = block->offset;
455 size_in_block = block->length;
458 if (phys_addr >= start && phys_addr < start + size_in_block) {
459 return phys_addr - start + offset;
462 offset += size_in_block;
465 return -1;
468 static int write_elf_loads(DumpState *s)
470 hwaddr offset;
471 MemoryMapping *memory_mapping;
472 uint32_t phdr_index = 1;
473 int ret;
474 uint32_t max_index;
476 if (s->have_section) {
477 max_index = s->sh_info;
478 } else {
479 max_index = s->phdr_num;
482 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
483 offset = get_offset(memory_mapping->phys_addr, s);
484 if (s->dump_info.d_class == ELFCLASS64) {
485 ret = write_elf64_load(s, memory_mapping, phdr_index++, offset);
486 } else {
487 ret = write_elf32_load(s, memory_mapping, phdr_index++, offset);
490 if (ret < 0) {
491 return -1;
494 if (phdr_index >= max_index) {
495 break;
499 return 0;
502 /* write elf header, PT_NOTE and elf note to vmcore. */
503 static int dump_begin(DumpState *s)
505 int ret;
508 * the vmcore's format is:
509 * --------------
510 * | elf header |
511 * --------------
512 * | PT_NOTE |
513 * --------------
514 * | PT_LOAD |
515 * --------------
516 * | ...... |
517 * --------------
518 * | PT_LOAD |
519 * --------------
520 * | sec_hdr |
521 * --------------
522 * | elf note |
523 * --------------
524 * | memory |
525 * --------------
527 * we only know where the memory is saved after we write elf note into
528 * vmcore.
531 /* write elf header to vmcore */
532 if (s->dump_info.d_class == ELFCLASS64) {
533 ret = write_elf64_header(s);
534 } else {
535 ret = write_elf32_header(s);
537 if (ret < 0) {
538 return -1;
541 if (s->dump_info.d_class == ELFCLASS64) {
542 /* write PT_NOTE to vmcore */
543 if (write_elf64_note(s) < 0) {
544 return -1;
547 /* write all PT_LOAD to vmcore */
548 if (write_elf_loads(s) < 0) {
549 return -1;
552 /* write section to vmcore */
553 if (s->have_section) {
554 if (write_elf_section(s, 1) < 0) {
555 return -1;
559 /* write notes to vmcore */
560 if (write_elf64_notes(s) < 0) {
561 return -1;
564 } else {
565 /* write PT_NOTE to vmcore */
566 if (write_elf32_note(s) < 0) {
567 return -1;
570 /* write all PT_LOAD to vmcore */
571 if (write_elf_loads(s) < 0) {
572 return -1;
575 /* write section to vmcore */
576 if (s->have_section) {
577 if (write_elf_section(s, 0) < 0) {
578 return -1;
582 /* write notes to vmcore */
583 if (write_elf32_notes(s) < 0) {
584 return -1;
588 return 0;
591 /* write PT_LOAD to vmcore */
592 static int dump_completed(DumpState *s)
594 dump_cleanup(s);
595 return 0;
598 static int get_next_block(DumpState *s, RAMBlock *block)
600 while (1) {
601 block = QTAILQ_NEXT(block, next);
602 if (!block) {
603 /* no more block */
604 return 1;
607 s->start = 0;
608 s->block = block;
609 if (s->has_filter) {
610 if (block->offset >= s->begin + s->length ||
611 block->offset + block->length <= s->begin) {
612 /* This block is out of the range */
613 continue;
616 if (s->begin > block->offset) {
617 s->start = s->begin - block->offset;
621 return 0;
625 /* write all memory to vmcore */
626 static int dump_iterate(DumpState *s)
628 RAMBlock *block;
629 int64_t size;
630 int ret;
632 while (1) {
633 block = s->block;
635 size = block->length;
636 if (s->has_filter) {
637 size -= s->start;
638 if (s->begin + s->length < block->offset + block->length) {
639 size -= block->offset + block->length - (s->begin + s->length);
642 ret = write_memory(s, block, s->start, size);
643 if (ret == -1) {
644 return ret;
647 ret = get_next_block(s, block);
648 if (ret == 1) {
649 dump_completed(s);
650 return 0;
655 static int create_vmcore(DumpState *s)
657 int ret;
659 ret = dump_begin(s);
660 if (ret < 0) {
661 return -1;
664 ret = dump_iterate(s);
665 if (ret < 0) {
666 return -1;
669 return 0;
672 static ram_addr_t get_start_block(DumpState *s)
674 RAMBlock *block;
676 if (!s->has_filter) {
677 s->block = QTAILQ_FIRST(&ram_list.blocks);
678 return 0;
681 QTAILQ_FOREACH(block, &ram_list.blocks, next) {
682 if (block->offset >= s->begin + s->length ||
683 block->offset + block->length <= s->begin) {
684 /* This block is out of the range */
685 continue;
688 s->block = block;
689 if (s->begin > block->offset) {
690 s->start = s->begin - block->offset;
691 } else {
692 s->start = 0;
694 return s->start;
697 return -1;
700 static int dump_init(DumpState *s, int fd, bool paging, bool has_filter,
701 int64_t begin, int64_t length, Error **errp)
703 CPUArchState *env;
704 int nr_cpus;
705 int ret;
707 if (runstate_is_running()) {
708 vm_stop(RUN_STATE_SAVE_VM);
709 s->resume = true;
710 } else {
711 s->resume = false;
714 s->errp = errp;
715 s->fd = fd;
716 s->has_filter = has_filter;
717 s->begin = begin;
718 s->length = length;
719 s->start = get_start_block(s);
720 if (s->start == -1) {
721 error_set(errp, QERR_INVALID_PARAMETER, "begin");
722 goto cleanup;
726 * get dump info: endian, class and architecture.
727 * If the target architecture is not supported, cpu_get_dump_info() will
728 * return -1.
730 * if we use kvm, we should synchronize the register before we get dump
731 * info.
733 nr_cpus = 0;
734 for (env = first_cpu; env != NULL; env = env->next_cpu) {
735 cpu_synchronize_state(env);
736 nr_cpus++;
739 ret = cpu_get_dump_info(&s->dump_info);
740 if (ret < 0) {
741 error_set(errp, QERR_UNSUPPORTED);
742 goto cleanup;
745 s->note_size = cpu_get_note_size(s->dump_info.d_class,
746 s->dump_info.d_machine, nr_cpus);
747 if (ret < 0) {
748 error_set(errp, QERR_UNSUPPORTED);
749 goto cleanup;
752 /* get memory mapping */
753 memory_mapping_list_init(&s->list);
754 if (paging) {
755 qemu_get_guest_memory_mapping(&s->list);
756 } else {
757 qemu_get_guest_simple_memory_mapping(&s->list);
760 if (s->has_filter) {
761 memory_mapping_filter(&s->list, s->begin, s->length);
765 * calculate phdr_num
767 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
769 s->phdr_num = 1; /* PT_NOTE */
770 if (s->list.num < UINT16_MAX - 2) {
771 s->phdr_num += s->list.num;
772 s->have_section = false;
773 } else {
774 s->have_section = true;
775 s->phdr_num = PN_XNUM;
776 s->sh_info = 1; /* PT_NOTE */
778 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
779 if (s->list.num <= UINT32_MAX - 1) {
780 s->sh_info += s->list.num;
781 } else {
782 s->sh_info = UINT32_MAX;
786 if (s->dump_info.d_class == ELFCLASS64) {
787 if (s->have_section) {
788 s->memory_offset = sizeof(Elf64_Ehdr) +
789 sizeof(Elf64_Phdr) * s->sh_info +
790 sizeof(Elf64_Shdr) + s->note_size;
791 } else {
792 s->memory_offset = sizeof(Elf64_Ehdr) +
793 sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
795 } else {
796 if (s->have_section) {
797 s->memory_offset = sizeof(Elf32_Ehdr) +
798 sizeof(Elf32_Phdr) * s->sh_info +
799 sizeof(Elf32_Shdr) + s->note_size;
800 } else {
801 s->memory_offset = sizeof(Elf32_Ehdr) +
802 sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
806 return 0;
808 cleanup:
809 if (s->resume) {
810 vm_start();
813 return -1;
816 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
817 int64_t begin, bool has_length, int64_t length,
818 Error **errp)
820 const char *p;
821 int fd = -1;
822 DumpState *s;
823 int ret;
825 if (has_begin && !has_length) {
826 error_set(errp, QERR_MISSING_PARAMETER, "length");
827 return;
829 if (!has_begin && has_length) {
830 error_set(errp, QERR_MISSING_PARAMETER, "begin");
831 return;
834 #if !defined(WIN32)
835 if (strstart(file, "fd:", &p)) {
836 fd = monitor_get_fd(cur_mon, p, errp);
837 if (fd == -1) {
838 return;
841 #endif
843 if (strstart(file, "file:", &p)) {
844 fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
845 if (fd < 0) {
846 error_set(errp, QERR_OPEN_FILE_FAILED, p);
847 return;
851 if (fd == -1) {
852 error_set(errp, QERR_INVALID_PARAMETER, "protocol");
853 return;
856 s = g_malloc(sizeof(DumpState));
858 ret = dump_init(s, fd, paging, has_begin, begin, length, errp);
859 if (ret < 0) {
860 g_free(s);
861 return;
864 if (create_vmcore(s) < 0 && !error_is_set(s->errp)) {
865 error_set(errp, QERR_IO_ERROR);
868 g_free(s);