xref: /linux/kernel/debug/kdb/kdb_support.c (revision a1c3be890440a1769ed6f822376a3e3ab0d42994)
1 /*
2  * Kernel Debugger Architecture Independent Support Functions
3  *
4  * This file is subject to the terms and conditions of the GNU General Public
5  * License.  See the file "COPYING" in the main directory of this archive
6  * for more details.
7  *
8  * Copyright (c) 1999-2004 Silicon Graphics, Inc.  All Rights Reserved.
9  * Copyright (c) 2009 Wind River Systems, Inc.  All Rights Reserved.
10  * 03/02/13    added new 2.5 kallsyms <xavier.bru@bull.net>
11  */
12 
13 #include <stdarg.h>
14 #include <linux/types.h>
15 #include <linux/sched.h>
16 #include <linux/mm.h>
17 #include <linux/kallsyms.h>
18 #include <linux/stddef.h>
19 #include <linux/vmalloc.h>
20 #include <linux/ptrace.h>
21 #include <linux/module.h>
22 #include <linux/highmem.h>
23 #include <linux/hardirq.h>
24 #include <linux/delay.h>
25 #include <linux/uaccess.h>
26 #include <linux/kdb.h>
27 #include <linux/slab.h>
28 #include "kdb_private.h"
29 
30 /*
31  * kdbgetsymval - Return the address of the given symbol.
32  *
33  * Parameters:
34  *	symname	Character string containing symbol name
35  *      symtab  Structure to receive results
36  * Returns:
37  *	0	Symbol not found, symtab zero filled
38  *	1	Symbol mapped to module/symbol/section, data in symtab
39  */
40 int kdbgetsymval(const char *symname, kdb_symtab_t *symtab)
41 {
42 	kdb_dbg_printf(AR, "symname=%s, symtab=%px\n", symname, symtab);
43 	memset(symtab, 0, sizeof(*symtab));
44 	symtab->sym_start = kallsyms_lookup_name(symname);
45 	if (symtab->sym_start) {
46 		kdb_dbg_printf(AR, "returns 1, symtab->sym_start=0x%lx\n",
47 			       symtab->sym_start);
48 		return 1;
49 	}
50 	kdb_dbg_printf(AR, "returns 0\n");
51 	return 0;
52 }
53 EXPORT_SYMBOL(kdbgetsymval);
54 
55 static char *kdb_name_table[100];	/* arbitrary size */
56 
57 /*
58  * kdbnearsym -	Return the name of the symbol with the nearest address
59  *	less than 'addr'.
60  *
61  * Parameters:
62  *	addr	Address to check for symbol near
63  *	symtab  Structure to receive results
64  * Returns:
65  *	0	No sections contain this address, symtab zero filled
66  *	1	Address mapped to module/symbol/section, data in symtab
67  * Remarks:
68  *	2.6 kallsyms has a "feature" where it unpacks the name into a
69  *	string.  If that string is reused before the caller expects it
70  *	then the caller sees its string change without warning.  To
71  *	avoid cluttering up the main kdb code with lots of kdb_strdup,
72  *	tests and kfree calls, kdbnearsym maintains an LRU list of the
73  *	last few unique strings.  The list is sized large enough to
74  *	hold active strings, no kdb caller of kdbnearsym makes more
75  *	than ~20 later calls before using a saved value.
76  */
77 int kdbnearsym(unsigned long addr, kdb_symtab_t *symtab)
78 {
79 	int ret = 0;
80 	unsigned long symbolsize = 0;
81 	unsigned long offset = 0;
82 #define knt1_size 128		/* must be >= kallsyms table size */
83 	char *knt1 = NULL;
84 
85 	kdb_dbg_printf(AR, "addr=0x%lx, symtab=%px\n", addr, symtab);
86 	memset(symtab, 0, sizeof(*symtab));
87 
88 	if (addr < 4096)
89 		goto out;
90 	knt1 = debug_kmalloc(knt1_size, GFP_ATOMIC);
91 	if (!knt1) {
92 		kdb_func_printf("addr=0x%lx cannot kmalloc knt1\n", addr);
93 		goto out;
94 	}
95 	symtab->sym_name = kallsyms_lookup(addr, &symbolsize , &offset,
96 				(char **)(&symtab->mod_name), knt1);
97 	if (offset > 8*1024*1024) {
98 		symtab->sym_name = NULL;
99 		addr = offset = symbolsize = 0;
100 	}
101 	symtab->sym_start = addr - offset;
102 	symtab->sym_end = symtab->sym_start + symbolsize;
103 	ret = symtab->sym_name != NULL && *(symtab->sym_name) != '\0';
104 
105 	if (ret) {
106 		int i;
107 		/* Another 2.6 kallsyms "feature".  Sometimes the sym_name is
108 		 * set but the buffer passed into kallsyms_lookup is not used,
109 		 * so it contains garbage.  The caller has to work out which
110 		 * buffer needs to be saved.
111 		 *
112 		 * What was Rusty smoking when he wrote that code?
113 		 */
114 		if (symtab->sym_name != knt1) {
115 			strncpy(knt1, symtab->sym_name, knt1_size);
116 			knt1[knt1_size-1] = '\0';
117 		}
118 		for (i = 0; i < ARRAY_SIZE(kdb_name_table); ++i) {
119 			if (kdb_name_table[i] &&
120 			    strcmp(kdb_name_table[i], knt1) == 0)
121 				break;
122 		}
123 		if (i >= ARRAY_SIZE(kdb_name_table)) {
124 			debug_kfree(kdb_name_table[0]);
125 			memmove(kdb_name_table, kdb_name_table+1,
126 			       sizeof(kdb_name_table[0]) *
127 			       (ARRAY_SIZE(kdb_name_table)-1));
128 		} else {
129 			debug_kfree(knt1);
130 			knt1 = kdb_name_table[i];
131 			memmove(kdb_name_table+i, kdb_name_table+i+1,
132 			       sizeof(kdb_name_table[0]) *
133 			       (ARRAY_SIZE(kdb_name_table)-i-1));
134 		}
135 		i = ARRAY_SIZE(kdb_name_table) - 1;
136 		kdb_name_table[i] = knt1;
137 		symtab->sym_name = kdb_name_table[i];
138 		knt1 = NULL;
139 	}
140 
141 	if (symtab->mod_name == NULL)
142 		symtab->mod_name = "kernel";
143 	kdb_dbg_printf(AR, "returns %d symtab->sym_start=0x%lx, symtab->mod_name=%px, symtab->sym_name=%px (%s)\n",
144 		       ret, symtab->sym_start, symtab->mod_name, symtab->sym_name, symtab->sym_name);
145 
146 out:
147 	debug_kfree(knt1);
148 	return ret;
149 }
150 
151 void kdbnearsym_cleanup(void)
152 {
153 	int i;
154 	for (i = 0; i < ARRAY_SIZE(kdb_name_table); ++i) {
155 		if (kdb_name_table[i]) {
156 			debug_kfree(kdb_name_table[i]);
157 			kdb_name_table[i] = NULL;
158 		}
159 	}
160 }
161 
162 static char ks_namebuf[KSYM_NAME_LEN+1], ks_namebuf_prev[KSYM_NAME_LEN+1];
163 
164 /*
165  * kallsyms_symbol_complete
166  *
167  * Parameters:
168  *	prefix_name	prefix of a symbol name to lookup
169  *	max_len		maximum length that can be returned
170  * Returns:
171  *	Number of symbols which match the given prefix.
172  * Notes:
173  *	prefix_name is changed to contain the longest unique prefix that
174  *	starts with this prefix (tab completion).
175  */
176 int kallsyms_symbol_complete(char *prefix_name, int max_len)
177 {
178 	loff_t pos = 0;
179 	int prefix_len = strlen(prefix_name), prev_len = 0;
180 	int i, number = 0;
181 	const char *name;
182 
183 	while ((name = kdb_walk_kallsyms(&pos))) {
184 		if (strncmp(name, prefix_name, prefix_len) == 0) {
185 			strscpy(ks_namebuf, name, sizeof(ks_namebuf));
186 			/* Work out the longest name that matches the prefix */
187 			if (++number == 1) {
188 				prev_len = min_t(int, max_len-1,
189 						 strlen(ks_namebuf));
190 				memcpy(ks_namebuf_prev, ks_namebuf, prev_len);
191 				ks_namebuf_prev[prev_len] = '\0';
192 				continue;
193 			}
194 			for (i = 0; i < prev_len; i++) {
195 				if (ks_namebuf[i] != ks_namebuf_prev[i]) {
196 					prev_len = i;
197 					ks_namebuf_prev[i] = '\0';
198 					break;
199 				}
200 			}
201 		}
202 	}
203 	if (prev_len > prefix_len)
204 		memcpy(prefix_name, ks_namebuf_prev, prev_len+1);
205 	return number;
206 }
207 
208 /*
209  * kallsyms_symbol_next
210  *
211  * Parameters:
212  *	prefix_name	prefix of a symbol name to lookup
213  *	flag	0 means search from the head, 1 means continue search.
214  *	buf_size	maximum length that can be written to prefix_name
215  *			buffer
216  * Returns:
217  *	1 if a symbol matches the given prefix.
218  *	0 if no string found
219  */
220 int kallsyms_symbol_next(char *prefix_name, int flag, int buf_size)
221 {
222 	int prefix_len = strlen(prefix_name);
223 	static loff_t pos;
224 	const char *name;
225 
226 	if (!flag)
227 		pos = 0;
228 
229 	while ((name = kdb_walk_kallsyms(&pos))) {
230 		if (!strncmp(name, prefix_name, prefix_len))
231 			return strscpy(prefix_name, name, buf_size);
232 	}
233 	return 0;
234 }
235 
236 /*
237  * kdb_symbol_print - Standard method for printing a symbol name and offset.
238  * Inputs:
239  *	addr	Address to be printed.
240  *	symtab	Address of symbol data, if NULL this routine does its
241  *		own lookup.
242  *	punc	Punctuation for string, bit field.
243  * Remarks:
244  *	The string and its punctuation is only printed if the address
245  *	is inside the kernel, except that the value is always printed
246  *	when requested.
247  */
248 void kdb_symbol_print(unsigned long addr, const kdb_symtab_t *symtab_p,
249 		      unsigned int punc)
250 {
251 	kdb_symtab_t symtab, *symtab_p2;
252 	if (symtab_p) {
253 		symtab_p2 = (kdb_symtab_t *)symtab_p;
254 	} else {
255 		symtab_p2 = &symtab;
256 		kdbnearsym(addr, symtab_p2);
257 	}
258 	if (!(symtab_p2->sym_name || (punc & KDB_SP_VALUE)))
259 		return;
260 	if (punc & KDB_SP_SPACEB)
261 		kdb_printf(" ");
262 	if (punc & KDB_SP_VALUE)
263 		kdb_printf(kdb_machreg_fmt0, addr);
264 	if (symtab_p2->sym_name) {
265 		if (punc & KDB_SP_VALUE)
266 			kdb_printf(" ");
267 		if (punc & KDB_SP_PAREN)
268 			kdb_printf("(");
269 		if (strcmp(symtab_p2->mod_name, "kernel"))
270 			kdb_printf("[%s]", symtab_p2->mod_name);
271 		kdb_printf("%s", symtab_p2->sym_name);
272 		if (addr != symtab_p2->sym_start)
273 			kdb_printf("+0x%lx", addr - symtab_p2->sym_start);
274 		if (punc & KDB_SP_SYMSIZE)
275 			kdb_printf("/0x%lx",
276 				   symtab_p2->sym_end - symtab_p2->sym_start);
277 		if (punc & KDB_SP_PAREN)
278 			kdb_printf(")");
279 	}
280 	if (punc & KDB_SP_SPACEA)
281 		kdb_printf(" ");
282 	if (punc & KDB_SP_NEWLINE)
283 		kdb_printf("\n");
284 }
285 
286 /*
287  * kdb_strdup - kdb equivalent of strdup, for disasm code.
288  * Inputs:
289  *	str	The string to duplicate.
290  *	type	Flags to kmalloc for the new string.
291  * Returns:
292  *	Address of the new string, NULL if storage could not be allocated.
293  * Remarks:
294  *	This is not in lib/string.c because it uses kmalloc which is not
295  *	available when string.o is used in boot loaders.
296  */
297 char *kdb_strdup(const char *str, gfp_t type)
298 {
299 	int n = strlen(str)+1;
300 	char *s = kmalloc(n, type);
301 	if (!s)
302 		return NULL;
303 	return strcpy(s, str);
304 }
305 
306 /*
307  * kdb_getarea_size - Read an area of data.  The kdb equivalent of
308  *	copy_from_user, with kdb messages for invalid addresses.
309  * Inputs:
310  *	res	Pointer to the area to receive the result.
311  *	addr	Address of the area to copy.
312  *	size	Size of the area.
313  * Returns:
314  *	0 for success, < 0 for error.
315  */
316 int kdb_getarea_size(void *res, unsigned long addr, size_t size)
317 {
318 	int ret = copy_from_kernel_nofault((char *)res, (char *)addr, size);
319 	if (ret) {
320 		if (!KDB_STATE(SUPPRESS)) {
321 			kdb_func_printf("Bad address 0x%lx\n", addr);
322 			KDB_STATE_SET(SUPPRESS);
323 		}
324 		ret = KDB_BADADDR;
325 	} else {
326 		KDB_STATE_CLEAR(SUPPRESS);
327 	}
328 	return ret;
329 }
330 
331 /*
332  * kdb_putarea_size - Write an area of data.  The kdb equivalent of
333  *	copy_to_user, with kdb messages for invalid addresses.
334  * Inputs:
335  *	addr	Address of the area to write to.
336  *	res	Pointer to the area holding the data.
337  *	size	Size of the area.
338  * Returns:
339  *	0 for success, < 0 for error.
340  */
341 int kdb_putarea_size(unsigned long addr, void *res, size_t size)
342 {
343 	int ret = copy_from_kernel_nofault((char *)addr, (char *)res, size);
344 	if (ret) {
345 		if (!KDB_STATE(SUPPRESS)) {
346 			kdb_func_printf("Bad address 0x%lx\n", addr);
347 			KDB_STATE_SET(SUPPRESS);
348 		}
349 		ret = KDB_BADADDR;
350 	} else {
351 		KDB_STATE_CLEAR(SUPPRESS);
352 	}
353 	return ret;
354 }
355 
356 /*
357  * kdb_getphys - Read data from a physical address. Validate the
358  * 	address is in range, use kmap_atomic() to get data
359  * 	similar to kdb_getarea() - but for phys addresses
360  * Inputs:
361  * 	res	Pointer to the word to receive the result
362  * 	addr	Physical address of the area to copy
363  * 	size	Size of the area
364  * Returns:
365  *	0 for success, < 0 for error.
366  */
367 static int kdb_getphys(void *res, unsigned long addr, size_t size)
368 {
369 	unsigned long pfn;
370 	void *vaddr;
371 	struct page *page;
372 
373 	pfn = (addr >> PAGE_SHIFT);
374 	if (!pfn_valid(pfn))
375 		return 1;
376 	page = pfn_to_page(pfn);
377 	vaddr = kmap_atomic(page);
378 	memcpy(res, vaddr + (addr & (PAGE_SIZE - 1)), size);
379 	kunmap_atomic(vaddr);
380 
381 	return 0;
382 }
383 
384 /*
385  * kdb_getphysword
386  * Inputs:
387  *	word	Pointer to the word to receive the result.
388  *	addr	Address of the area to copy.
389  *	size	Size of the area.
390  * Returns:
391  *	0 for success, < 0 for error.
392  */
393 int kdb_getphysword(unsigned long *word, unsigned long addr, size_t size)
394 {
395 	int diag;
396 	__u8  w1;
397 	__u16 w2;
398 	__u32 w4;
399 	__u64 w8;
400 	*word = 0;	/* Default value if addr or size is invalid */
401 
402 	switch (size) {
403 	case 1:
404 		diag = kdb_getphys(&w1, addr, sizeof(w1));
405 		if (!diag)
406 			*word = w1;
407 		break;
408 	case 2:
409 		diag = kdb_getphys(&w2, addr, sizeof(w2));
410 		if (!diag)
411 			*word = w2;
412 		break;
413 	case 4:
414 		diag = kdb_getphys(&w4, addr, sizeof(w4));
415 		if (!diag)
416 			*word = w4;
417 		break;
418 	case 8:
419 		if (size <= sizeof(*word)) {
420 			diag = kdb_getphys(&w8, addr, sizeof(w8));
421 			if (!diag)
422 				*word = w8;
423 			break;
424 		}
425 		fallthrough;
426 	default:
427 		diag = KDB_BADWIDTH;
428 		kdb_func_printf("bad width %zu\n", size);
429 	}
430 	return diag;
431 }
432 
433 /*
434  * kdb_getword - Read a binary value.  Unlike kdb_getarea, this treats
435  *	data as numbers.
436  * Inputs:
437  *	word	Pointer to the word to receive the result.
438  *	addr	Address of the area to copy.
439  *	size	Size of the area.
440  * Returns:
441  *	0 for success, < 0 for error.
442  */
443 int kdb_getword(unsigned long *word, unsigned long addr, size_t size)
444 {
445 	int diag;
446 	__u8  w1;
447 	__u16 w2;
448 	__u32 w4;
449 	__u64 w8;
450 	*word = 0;	/* Default value if addr or size is invalid */
451 	switch (size) {
452 	case 1:
453 		diag = kdb_getarea(w1, addr);
454 		if (!diag)
455 			*word = w1;
456 		break;
457 	case 2:
458 		diag = kdb_getarea(w2, addr);
459 		if (!diag)
460 			*word = w2;
461 		break;
462 	case 4:
463 		diag = kdb_getarea(w4, addr);
464 		if (!diag)
465 			*word = w4;
466 		break;
467 	case 8:
468 		if (size <= sizeof(*word)) {
469 			diag = kdb_getarea(w8, addr);
470 			if (!diag)
471 				*word = w8;
472 			break;
473 		}
474 		fallthrough;
475 	default:
476 		diag = KDB_BADWIDTH;
477 		kdb_func_printf("bad width %zu\n", size);
478 	}
479 	return diag;
480 }
481 
482 /*
483  * kdb_putword - Write a binary value.  Unlike kdb_putarea, this
484  *	treats data as numbers.
485  * Inputs:
486  *	addr	Address of the area to write to..
487  *	word	The value to set.
488  *	size	Size of the area.
489  * Returns:
490  *	0 for success, < 0 for error.
491  */
492 int kdb_putword(unsigned long addr, unsigned long word, size_t size)
493 {
494 	int diag;
495 	__u8  w1;
496 	__u16 w2;
497 	__u32 w4;
498 	__u64 w8;
499 	switch (size) {
500 	case 1:
501 		w1 = word;
502 		diag = kdb_putarea(addr, w1);
503 		break;
504 	case 2:
505 		w2 = word;
506 		diag = kdb_putarea(addr, w2);
507 		break;
508 	case 4:
509 		w4 = word;
510 		diag = kdb_putarea(addr, w4);
511 		break;
512 	case 8:
513 		if (size <= sizeof(word)) {
514 			w8 = word;
515 			diag = kdb_putarea(addr, w8);
516 			break;
517 		}
518 		fallthrough;
519 	default:
520 		diag = KDB_BADWIDTH;
521 		kdb_func_printf("bad width %zu\n", size);
522 	}
523 	return diag;
524 }
525 
526 /*
527  * kdb_task_state_string - Convert a string containing any of the
528  *	letters DRSTCZEUIMA to a mask for the process state field and
529  *	return the value.  If no argument is supplied, return the mask
530  *	that corresponds to environment variable PS, DRSTCZEU by
531  *	default.
532  * Inputs:
533  *	s	String to convert
534  * Returns:
535  *	Mask for process state.
536  * Notes:
537  *	The mask folds data from several sources into a single long value, so
538  *	be careful not to overlap the bits.  TASK_* bits are in the LSB,
539  *	special cases like UNRUNNABLE are in the MSB.  As of 2.6.10-rc1 there
540  *	is no overlap between TASK_* and EXIT_* but that may not always be
541  *	true, so EXIT_* bits are shifted left 16 bits before being stored in
542  *	the mask.
543  */
544 
545 /* unrunnable is < 0 */
546 #define UNRUNNABLE	(1UL << (8*sizeof(unsigned long) - 1))
547 #define RUNNING		(1UL << (8*sizeof(unsigned long) - 2))
548 #define IDLE		(1UL << (8*sizeof(unsigned long) - 3))
549 #define DAEMON		(1UL << (8*sizeof(unsigned long) - 4))
550 
551 unsigned long kdb_task_state_string(const char *s)
552 {
553 	long res = 0;
554 	if (!s) {
555 		s = kdbgetenv("PS");
556 		if (!s)
557 			s = "DRSTCZEU";	/* default value for ps */
558 	}
559 	while (*s) {
560 		switch (*s) {
561 		case 'D':
562 			res |= TASK_UNINTERRUPTIBLE;
563 			break;
564 		case 'R':
565 			res |= RUNNING;
566 			break;
567 		case 'S':
568 			res |= TASK_INTERRUPTIBLE;
569 			break;
570 		case 'T':
571 			res |= TASK_STOPPED;
572 			break;
573 		case 'C':
574 			res |= TASK_TRACED;
575 			break;
576 		case 'Z':
577 			res |= EXIT_ZOMBIE << 16;
578 			break;
579 		case 'E':
580 			res |= EXIT_DEAD << 16;
581 			break;
582 		case 'U':
583 			res |= UNRUNNABLE;
584 			break;
585 		case 'I':
586 			res |= IDLE;
587 			break;
588 		case 'M':
589 			res |= DAEMON;
590 			break;
591 		case 'A':
592 			res = ~0UL;
593 			break;
594 		default:
595 			  kdb_func_printf("unknown flag '%c' ignored\n", *s);
596 			  break;
597 		}
598 		++s;
599 	}
600 	return res;
601 }
602 
603 /*
604  * kdb_task_state_char - Return the character that represents the task state.
605  * Inputs:
606  *	p	struct task for the process
607  * Returns:
608  *	One character to represent the task state.
609  */
610 char kdb_task_state_char (const struct task_struct *p)
611 {
612 	int cpu;
613 	char state;
614 	unsigned long tmp;
615 
616 	if (!p ||
617 	    copy_from_kernel_nofault(&tmp, (char *)p, sizeof(unsigned long)))
618 		return 'E';
619 
620 	cpu = kdb_process_cpu(p);
621 	state = (p->state == 0) ? 'R' :
622 		(p->state < 0) ? 'U' :
623 		(p->state & TASK_UNINTERRUPTIBLE) ? 'D' :
624 		(p->state & TASK_STOPPED) ? 'T' :
625 		(p->state & TASK_TRACED) ? 'C' :
626 		(p->exit_state & EXIT_ZOMBIE) ? 'Z' :
627 		(p->exit_state & EXIT_DEAD) ? 'E' :
628 		(p->state & TASK_INTERRUPTIBLE) ? 'S' : '?';
629 	if (is_idle_task(p)) {
630 		/* Idle task.  Is it really idle, apart from the kdb
631 		 * interrupt? */
632 		if (!kdb_task_has_cpu(p) || kgdb_info[cpu].irq_depth == 1) {
633 			if (cpu != kdb_initial_cpu)
634 				state = 'I';	/* idle task */
635 		}
636 	} else if (!p->mm && state == 'S') {
637 		state = 'M';	/* sleeping system daemon */
638 	}
639 	return state;
640 }
641 
642 /*
643  * kdb_task_state - Return true if a process has the desired state
644  *	given by the mask.
645  * Inputs:
646  *	p	struct task for the process
647  *	mask	mask from kdb_task_state_string to select processes
648  * Returns:
649  *	True if the process matches at least one criteria defined by the mask.
650  */
651 unsigned long kdb_task_state(const struct task_struct *p, unsigned long mask)
652 {
653 	char state[] = { kdb_task_state_char(p), '\0' };
654 	return (mask & kdb_task_state_string(state)) != 0;
655 }
656 
657 /*
658  * kdb_print_nameval - Print a name and its value, converting the
659  *	value to a symbol lookup if possible.
660  * Inputs:
661  *	name	field name to print
662  *	val	value of field
663  */
664 void kdb_print_nameval(const char *name, unsigned long val)
665 {
666 	kdb_symtab_t symtab;
667 	kdb_printf("  %-11.11s ", name);
668 	if (kdbnearsym(val, &symtab))
669 		kdb_symbol_print(val, &symtab,
670 				 KDB_SP_VALUE|KDB_SP_SYMSIZE|KDB_SP_NEWLINE);
671 	else
672 		kdb_printf("0x%lx\n", val);
673 }
674 
675 /* Last ditch allocator for debugging, so we can still debug even when
676  * the GFP_ATOMIC pool has been exhausted.  The algorithms are tuned
677  * for space usage, not for speed.  One smallish memory pool, the free
678  * chain is always in ascending address order to allow coalescing,
679  * allocations are done in brute force best fit.
680  */
681 
682 struct debug_alloc_header {
683 	u32 next;	/* offset of next header from start of pool */
684 	u32 size;
685 	void *caller;
686 };
687 
688 /* The memory returned by this allocator must be aligned, which means
689  * so must the header size.  Do not assume that sizeof(struct
690  * debug_alloc_header) is a multiple of the alignment, explicitly
691  * calculate the overhead of this header, including the alignment.
692  * The rest of this code must not use sizeof() on any header or
693  * pointer to a header.
694  */
695 #define dah_align 8
696 #define dah_overhead ALIGN(sizeof(struct debug_alloc_header), dah_align)
697 
698 static u64 debug_alloc_pool_aligned[256*1024/dah_align];	/* 256K pool */
699 static char *debug_alloc_pool = (char *)debug_alloc_pool_aligned;
700 static u32 dah_first, dah_first_call = 1, dah_used, dah_used_max;
701 
702 /* Locking is awkward.  The debug code is called from all contexts,
703  * including non maskable interrupts.  A normal spinlock is not safe
704  * in NMI context.  Try to get the debug allocator lock, if it cannot
705  * be obtained after a second then give up.  If the lock could not be
706  * previously obtained on this cpu then only try once.
707  *
708  * sparse has no annotation for "this function _sometimes_ acquires a
709  * lock", so fudge the acquire/release notation.
710  */
711 static DEFINE_SPINLOCK(dap_lock);
712 static int get_dap_lock(void)
713 	__acquires(dap_lock)
714 {
715 	static int dap_locked = -1;
716 	int count;
717 	if (dap_locked == smp_processor_id())
718 		count = 1;
719 	else
720 		count = 1000;
721 	while (1) {
722 		if (spin_trylock(&dap_lock)) {
723 			dap_locked = -1;
724 			return 1;
725 		}
726 		if (!count--)
727 			break;
728 		udelay(1000);
729 	}
730 	dap_locked = smp_processor_id();
731 	__acquire(dap_lock);
732 	return 0;
733 }
734 
735 void *debug_kmalloc(size_t size, gfp_t flags)
736 {
737 	unsigned int rem, h_offset;
738 	struct debug_alloc_header *best, *bestprev, *prev, *h;
739 	void *p = NULL;
740 	if (!get_dap_lock()) {
741 		__release(dap_lock);	/* we never actually got it */
742 		return NULL;
743 	}
744 	h = (struct debug_alloc_header *)(debug_alloc_pool + dah_first);
745 	if (dah_first_call) {
746 		h->size = sizeof(debug_alloc_pool_aligned) - dah_overhead;
747 		dah_first_call = 0;
748 	}
749 	size = ALIGN(size, dah_align);
750 	prev = best = bestprev = NULL;
751 	while (1) {
752 		if (h->size >= size && (!best || h->size < best->size)) {
753 			best = h;
754 			bestprev = prev;
755 			if (h->size == size)
756 				break;
757 		}
758 		if (!h->next)
759 			break;
760 		prev = h;
761 		h = (struct debug_alloc_header *)(debug_alloc_pool + h->next);
762 	}
763 	if (!best)
764 		goto out;
765 	rem = best->size - size;
766 	/* The pool must always contain at least one header */
767 	if (best->next == 0 && bestprev == NULL && rem < dah_overhead)
768 		goto out;
769 	if (rem >= dah_overhead) {
770 		best->size = size;
771 		h_offset = ((char *)best - debug_alloc_pool) +
772 			   dah_overhead + best->size;
773 		h = (struct debug_alloc_header *)(debug_alloc_pool + h_offset);
774 		h->size = rem - dah_overhead;
775 		h->next = best->next;
776 	} else
777 		h_offset = best->next;
778 	best->caller = __builtin_return_address(0);
779 	dah_used += best->size;
780 	dah_used_max = max(dah_used, dah_used_max);
781 	if (bestprev)
782 		bestprev->next = h_offset;
783 	else
784 		dah_first = h_offset;
785 	p = (char *)best + dah_overhead;
786 	memset(p, POISON_INUSE, best->size - 1);
787 	*((char *)p + best->size - 1) = POISON_END;
788 out:
789 	spin_unlock(&dap_lock);
790 	return p;
791 }
792 
793 void debug_kfree(void *p)
794 {
795 	struct debug_alloc_header *h;
796 	unsigned int h_offset;
797 	if (!p)
798 		return;
799 	if ((char *)p < debug_alloc_pool ||
800 	    (char *)p >= debug_alloc_pool + sizeof(debug_alloc_pool_aligned)) {
801 		kfree(p);
802 		return;
803 	}
804 	if (!get_dap_lock()) {
805 		__release(dap_lock);	/* we never actually got it */
806 		return;		/* memory leak, cannot be helped */
807 	}
808 	h = (struct debug_alloc_header *)((char *)p - dah_overhead);
809 	memset(p, POISON_FREE, h->size - 1);
810 	*((char *)p + h->size - 1) = POISON_END;
811 	h->caller = NULL;
812 	dah_used -= h->size;
813 	h_offset = (char *)h - debug_alloc_pool;
814 	if (h_offset < dah_first) {
815 		h->next = dah_first;
816 		dah_first = h_offset;
817 	} else {
818 		struct debug_alloc_header *prev;
819 		unsigned int prev_offset;
820 		prev = (struct debug_alloc_header *)(debug_alloc_pool +
821 						     dah_first);
822 		while (1) {
823 			if (!prev->next || prev->next > h_offset)
824 				break;
825 			prev = (struct debug_alloc_header *)
826 				(debug_alloc_pool + prev->next);
827 		}
828 		prev_offset = (char *)prev - debug_alloc_pool;
829 		if (prev_offset + dah_overhead + prev->size == h_offset) {
830 			prev->size += dah_overhead + h->size;
831 			memset(h, POISON_FREE, dah_overhead - 1);
832 			*((char *)h + dah_overhead - 1) = POISON_END;
833 			h = prev;
834 			h_offset = prev_offset;
835 		} else {
836 			h->next = prev->next;
837 			prev->next = h_offset;
838 		}
839 	}
840 	if (h_offset + dah_overhead + h->size == h->next) {
841 		struct debug_alloc_header *next;
842 		next = (struct debug_alloc_header *)
843 			(debug_alloc_pool + h->next);
844 		h->size += dah_overhead + next->size;
845 		h->next = next->next;
846 		memset(next, POISON_FREE, dah_overhead - 1);
847 		*((char *)next + dah_overhead - 1) = POISON_END;
848 	}
849 	spin_unlock(&dap_lock);
850 }
851 
852 void debug_kusage(void)
853 {
854 	struct debug_alloc_header *h_free, *h_used;
855 #ifdef	CONFIG_IA64
856 	/* FIXME: using dah for ia64 unwind always results in a memory leak.
857 	 * Fix that memory leak first, then set debug_kusage_one_time = 1 for
858 	 * all architectures.
859 	 */
860 	static int debug_kusage_one_time;
861 #else
862 	static int debug_kusage_one_time = 1;
863 #endif
864 	if (!get_dap_lock()) {
865 		__release(dap_lock);	/* we never actually got it */
866 		return;
867 	}
868 	h_free = (struct debug_alloc_header *)(debug_alloc_pool + dah_first);
869 	if (dah_first == 0 &&
870 	    (h_free->size == sizeof(debug_alloc_pool_aligned) - dah_overhead ||
871 	     dah_first_call))
872 		goto out;
873 	if (!debug_kusage_one_time)
874 		goto out;
875 	debug_kusage_one_time = 0;
876 	kdb_func_printf("debug_kmalloc memory leak dah_first %d\n", dah_first);
877 	if (dah_first) {
878 		h_used = (struct debug_alloc_header *)debug_alloc_pool;
879 		kdb_func_printf("h_used %px size %d\n", h_used, h_used->size);
880 	}
881 	do {
882 		h_used = (struct debug_alloc_header *)
883 			  ((char *)h_free + dah_overhead + h_free->size);
884 		kdb_func_printf("h_used %px size %d caller %px\n",
885 				h_used, h_used->size, h_used->caller);
886 		h_free = (struct debug_alloc_header *)
887 			  (debug_alloc_pool + h_free->next);
888 	} while (h_free->next);
889 	h_used = (struct debug_alloc_header *)
890 		  ((char *)h_free + dah_overhead + h_free->size);
891 	if ((char *)h_used - debug_alloc_pool !=
892 	    sizeof(debug_alloc_pool_aligned))
893 		kdb_func_printf("h_used %px size %d caller %px\n",
894 				h_used, h_used->size, h_used->caller);
895 out:
896 	spin_unlock(&dap_lock);
897 }
898 
899 /* Maintain a small stack of kdb_flags to allow recursion without disturbing
900  * the global kdb state.
901  */
902 
903 static int kdb_flags_stack[4], kdb_flags_index;
904 
905 void kdb_save_flags(void)
906 {
907 	BUG_ON(kdb_flags_index >= ARRAY_SIZE(kdb_flags_stack));
908 	kdb_flags_stack[kdb_flags_index++] = kdb_flags;
909 }
910 
911 void kdb_restore_flags(void)
912 {
913 	BUG_ON(kdb_flags_index <= 0);
914 	kdb_flags = kdb_flags_stack[--kdb_flags_index];
915 }
916