xref: /linux-6.15/kernel/debug/kdb/kdb_support.c (revision 39f75da7)
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 <[email protected]>
11  */
12 
13 #include <linux/types.h>
14 #include <linux/sched.h>
15 #include <linux/mm.h>
16 #include <linux/kallsyms.h>
17 #include <linux/stddef.h>
18 #include <linux/vmalloc.h>
19 #include <linux/ptrace.h>
20 #include <linux/module.h>
21 #include <linux/highmem.h>
22 #include <linux/hardirq.h>
23 #include <linux/delay.h>
24 #include <linux/uaccess.h>
25 #include <linux/kdb.h>
26 #include <linux/slab.h>
27 #include "kdb_private.h"
28 
29 /*
30  * kdbgetsymval - Return the address of the given symbol.
31  *
32  * Parameters:
33  *	symname	Character string containing symbol name
34  *      symtab  Structure to receive results
35  * Returns:
36  *	0	Symbol not found, symtab zero filled
37  *	1	Symbol mapped to module/symbol/section, data in symtab
38  */
39 int kdbgetsymval(const char *symname, kdb_symtab_t *symtab)
40 {
41 	kdb_dbg_printf(AR, "symname=%s, symtab=%px\n", symname, symtab);
42 	memset(symtab, 0, sizeof(*symtab));
43 	symtab->sym_start = kallsyms_lookup_name(symname);
44 	if (symtab->sym_start) {
45 		kdb_dbg_printf(AR, "returns 1, symtab->sym_start=0x%lx\n",
46 			       symtab->sym_start);
47 		return 1;
48 	}
49 	kdb_dbg_printf(AR, "returns 0\n");
50 	return 0;
51 }
52 EXPORT_SYMBOL(kdbgetsymval);
53 
54 static char *kdb_name_table[100];	/* arbitrary size */
55 
56 /*
57  * kdbnearsym -	Return the name of the symbol with the nearest address
58  *	less than 'addr'.
59  *
60  * Parameters:
61  *	addr	Address to check for symbol near
62  *	symtab  Structure to receive results
63  * Returns:
64  *	0	No sections contain this address, symtab zero filled
65  *	1	Address mapped to module/symbol/section, data in symtab
66  * Remarks:
67  *	2.6 kallsyms has a "feature" where it unpacks the name into a
68  *	string.  If that string is reused before the caller expects it
69  *	then the caller sees its string change without warning.  To
70  *	avoid cluttering up the main kdb code with lots of kdb_strdup,
71  *	tests and kfree calls, kdbnearsym maintains an LRU list of the
72  *	last few unique strings.  The list is sized large enough to
73  *	hold active strings, no kdb caller of kdbnearsym makes more
74  *	than ~20 later calls before using a saved value.
75  */
76 int kdbnearsym(unsigned long addr, kdb_symtab_t *symtab)
77 {
78 	int ret = 0;
79 	unsigned long symbolsize = 0;
80 	unsigned long offset = 0;
81 #define knt1_size 128		/* must be >= kallsyms table size */
82 	char *knt1 = NULL;
83 
84 	kdb_dbg_printf(AR, "addr=0x%lx, symtab=%px\n", addr, symtab);
85 	memset(symtab, 0, sizeof(*symtab));
86 
87 	if (addr < 4096)
88 		goto out;
89 	knt1 = debug_kmalloc(knt1_size, GFP_ATOMIC);
90 	if (!knt1) {
91 		kdb_func_printf("addr=0x%lx cannot kmalloc knt1\n", addr);
92 		goto out;
93 	}
94 	symtab->sym_name = kallsyms_lookup(addr, &symbolsize , &offset,
95 				(char **)(&symtab->mod_name), knt1);
96 	if (offset > 8*1024*1024) {
97 		symtab->sym_name = NULL;
98 		addr = offset = symbolsize = 0;
99 	}
100 	symtab->sym_start = addr - offset;
101 	symtab->sym_end = symtab->sym_start + symbolsize;
102 	ret = symtab->sym_name != NULL && *(symtab->sym_name) != '\0';
103 
104 	if (ret) {
105 		int i;
106 		/* Another 2.6 kallsyms "feature".  Sometimes the sym_name is
107 		 * set but the buffer passed into kallsyms_lookup is not used,
108 		 * so it contains garbage.  The caller has to work out which
109 		 * buffer needs to be saved.
110 		 *
111 		 * What was Rusty smoking when he wrote that code?
112 		 */
113 		if (symtab->sym_name != knt1) {
114 			strncpy(knt1, symtab->sym_name, knt1_size);
115 			knt1[knt1_size-1] = '\0';
116 		}
117 		for (i = 0; i < ARRAY_SIZE(kdb_name_table); ++i) {
118 			if (kdb_name_table[i] &&
119 			    strcmp(kdb_name_table[i], knt1) == 0)
120 				break;
121 		}
122 		if (i >= ARRAY_SIZE(kdb_name_table)) {
123 			debug_kfree(kdb_name_table[0]);
124 			memmove(kdb_name_table, kdb_name_table+1,
125 			       sizeof(kdb_name_table[0]) *
126 			       (ARRAY_SIZE(kdb_name_table)-1));
127 		} else {
128 			debug_kfree(knt1);
129 			knt1 = kdb_name_table[i];
130 			memmove(kdb_name_table+i, kdb_name_table+i+1,
131 			       sizeof(kdb_name_table[0]) *
132 			       (ARRAY_SIZE(kdb_name_table)-i-1));
133 		}
134 		i = ARRAY_SIZE(kdb_name_table) - 1;
135 		kdb_name_table[i] = knt1;
136 		symtab->sym_name = kdb_name_table[i];
137 		knt1 = NULL;
138 	}
139 
140 	if (symtab->mod_name == NULL)
141 		symtab->mod_name = "kernel";
142 	kdb_dbg_printf(AR, "returns %d symtab->sym_start=0x%lx, symtab->mod_name=%px, symtab->sym_name=%px (%s)\n",
143 		       ret, symtab->sym_start, symtab->mod_name, symtab->sym_name, symtab->sym_name);
144 
145 out:
146 	debug_kfree(knt1);
147 	return ret;
148 }
149 
150 void kdbnearsym_cleanup(void)
151 {
152 	int i;
153 	for (i = 0; i < ARRAY_SIZE(kdb_name_table); ++i) {
154 		if (kdb_name_table[i]) {
155 			debug_kfree(kdb_name_table[i]);
156 			kdb_name_table[i] = NULL;
157 		}
158 	}
159 }
160 
161 static char ks_namebuf[KSYM_NAME_LEN+1], ks_namebuf_prev[KSYM_NAME_LEN+1];
162 
163 /*
164  * kallsyms_symbol_complete
165  *
166  * Parameters:
167  *	prefix_name	prefix of a symbol name to lookup
168  *	max_len		maximum length that can be returned
169  * Returns:
170  *	Number of symbols which match the given prefix.
171  * Notes:
172  *	prefix_name is changed to contain the longest unique prefix that
173  *	starts with this prefix (tab completion).
174  */
175 int kallsyms_symbol_complete(char *prefix_name, int max_len)
176 {
177 	loff_t pos = 0;
178 	int prefix_len = strlen(prefix_name), prev_len = 0;
179 	int i, number = 0;
180 	const char *name;
181 
182 	while ((name = kdb_walk_kallsyms(&pos))) {
183 		if (strncmp(name, prefix_name, prefix_len) == 0) {
184 			strscpy(ks_namebuf, name, sizeof(ks_namebuf));
185 			/* Work out the longest name that matches the prefix */
186 			if (++number == 1) {
187 				prev_len = min_t(int, max_len-1,
188 						 strlen(ks_namebuf));
189 				memcpy(ks_namebuf_prev, ks_namebuf, prev_len);
190 				ks_namebuf_prev[prev_len] = '\0';
191 				continue;
192 			}
193 			for (i = 0; i < prev_len; i++) {
194 				if (ks_namebuf[i] != ks_namebuf_prev[i]) {
195 					prev_len = i;
196 					ks_namebuf_prev[i] = '\0';
197 					break;
198 				}
199 			}
200 		}
201 	}
202 	if (prev_len > prefix_len)
203 		memcpy(prefix_name, ks_namebuf_prev, prev_len+1);
204 	return number;
205 }
206 
207 /*
208  * kallsyms_symbol_next
209  *
210  * Parameters:
211  *	prefix_name	prefix of a symbol name to lookup
212  *	flag	0 means search from the head, 1 means continue search.
213  *	buf_size	maximum length that can be written to prefix_name
214  *			buffer
215  * Returns:
216  *	1 if a symbol matches the given prefix.
217  *	0 if no string found
218  */
219 int kallsyms_symbol_next(char *prefix_name, int flag, int buf_size)
220 {
221 	int prefix_len = strlen(prefix_name);
222 	static loff_t pos;
223 	const char *name;
224 
225 	if (!flag)
226 		pos = 0;
227 
228 	while ((name = kdb_walk_kallsyms(&pos))) {
229 		if (!strncmp(name, prefix_name, prefix_len))
230 			return strscpy(prefix_name, name, buf_size);
231 	}
232 	return 0;
233 }
234 
235 /*
236  * kdb_symbol_print - Standard method for printing a symbol name and offset.
237  * Inputs:
238  *	addr	Address to be printed.
239  *	symtab	Address of symbol data, if NULL this routine does its
240  *		own lookup.
241  *	punc	Punctuation for string, bit field.
242  * Remarks:
243  *	The string and its punctuation is only printed if the address
244  *	is inside the kernel, except that the value is always printed
245  *	when requested.
246  */
247 void kdb_symbol_print(unsigned long addr, const kdb_symtab_t *symtab_p,
248 		      unsigned int punc)
249 {
250 	kdb_symtab_t symtab, *symtab_p2;
251 	if (symtab_p) {
252 		symtab_p2 = (kdb_symtab_t *)symtab_p;
253 	} else {
254 		symtab_p2 = &symtab;
255 		kdbnearsym(addr, symtab_p2);
256 	}
257 	if (!(symtab_p2->sym_name || (punc & KDB_SP_VALUE)))
258 		return;
259 	if (punc & KDB_SP_SPACEB)
260 		kdb_printf(" ");
261 	if (punc & KDB_SP_VALUE)
262 		kdb_printf(kdb_machreg_fmt0, addr);
263 	if (symtab_p2->sym_name) {
264 		if (punc & KDB_SP_VALUE)
265 			kdb_printf(" ");
266 		if (punc & KDB_SP_PAREN)
267 			kdb_printf("(");
268 		if (strcmp(symtab_p2->mod_name, "kernel"))
269 			kdb_printf("[%s]", symtab_p2->mod_name);
270 		kdb_printf("%s", symtab_p2->sym_name);
271 		if (addr != symtab_p2->sym_start)
272 			kdb_printf("+0x%lx", addr - symtab_p2->sym_start);
273 		if (punc & KDB_SP_SYMSIZE)
274 			kdb_printf("/0x%lx",
275 				   symtab_p2->sym_end - symtab_p2->sym_start);
276 		if (punc & KDB_SP_PAREN)
277 			kdb_printf(")");
278 	}
279 	if (punc & KDB_SP_SPACEA)
280 		kdb_printf(" ");
281 	if (punc & KDB_SP_NEWLINE)
282 		kdb_printf("\n");
283 }
284 
285 /*
286  * kdb_strdup - kdb equivalent of strdup, for disasm code.
287  * Inputs:
288  *	str	The string to duplicate.
289  *	type	Flags to kmalloc for the new string.
290  * Returns:
291  *	Address of the new string, NULL if storage could not be allocated.
292  * Remarks:
293  *	This is not in lib/string.c because it uses kmalloc which is not
294  *	available when string.o is used in boot loaders.
295  */
296 char *kdb_strdup(const char *str, gfp_t type)
297 {
298 	int n = strlen(str)+1;
299 	char *s = kmalloc(n, type);
300 	if (!s)
301 		return NULL;
302 	return strcpy(s, str);
303 }
304 
305 /*
306  * kdb_getarea_size - Read an area of data.  The kdb equivalent of
307  *	copy_from_user, with kdb messages for invalid addresses.
308  * Inputs:
309  *	res	Pointer to the area to receive the result.
310  *	addr	Address of the area to copy.
311  *	size	Size of the area.
312  * Returns:
313  *	0 for success, < 0 for error.
314  */
315 int kdb_getarea_size(void *res, unsigned long addr, size_t size)
316 {
317 	int ret = copy_from_kernel_nofault((char *)res, (char *)addr, size);
318 	if (ret) {
319 		if (!KDB_STATE(SUPPRESS)) {
320 			kdb_func_printf("Bad address 0x%lx\n", addr);
321 			KDB_STATE_SET(SUPPRESS);
322 		}
323 		ret = KDB_BADADDR;
324 	} else {
325 		KDB_STATE_CLEAR(SUPPRESS);
326 	}
327 	return ret;
328 }
329 
330 /*
331  * kdb_putarea_size - Write an area of data.  The kdb equivalent of
332  *	copy_to_user, with kdb messages for invalid addresses.
333  * Inputs:
334  *	addr	Address of the area to write to.
335  *	res	Pointer to the area holding the data.
336  *	size	Size of the area.
337  * Returns:
338  *	0 for success, < 0 for error.
339  */
340 int kdb_putarea_size(unsigned long addr, void *res, size_t size)
341 {
342 	int ret = copy_from_kernel_nofault((char *)addr, (char *)res, size);
343 	if (ret) {
344 		if (!KDB_STATE(SUPPRESS)) {
345 			kdb_func_printf("Bad address 0x%lx\n", addr);
346 			KDB_STATE_SET(SUPPRESS);
347 		}
348 		ret = KDB_BADADDR;
349 	} else {
350 		KDB_STATE_CLEAR(SUPPRESS);
351 	}
352 	return ret;
353 }
354 
355 /*
356  * kdb_getphys - Read data from a physical address. Validate the
357  * 	address is in range, use kmap_atomic() to get data
358  * 	similar to kdb_getarea() - but for phys addresses
359  * Inputs:
360  * 	res	Pointer to the word to receive the result
361  * 	addr	Physical address of the area to copy
362  * 	size	Size of the area
363  * Returns:
364  *	0 for success, < 0 for error.
365  */
366 static int kdb_getphys(void *res, unsigned long addr, size_t size)
367 {
368 	unsigned long pfn;
369 	void *vaddr;
370 	struct page *page;
371 
372 	pfn = (addr >> PAGE_SHIFT);
373 	if (!pfn_valid(pfn))
374 		return 1;
375 	page = pfn_to_page(pfn);
376 	vaddr = kmap_atomic(page);
377 	memcpy(res, vaddr + (addr & (PAGE_SIZE - 1)), size);
378 	kunmap_atomic(vaddr);
379 
380 	return 0;
381 }
382 
383 /*
384  * kdb_getphysword
385  * Inputs:
386  *	word	Pointer to the word to receive the result.
387  *	addr	Address of the area to copy.
388  *	size	Size of the area.
389  * Returns:
390  *	0 for success, < 0 for error.
391  */
392 int kdb_getphysword(unsigned long *word, unsigned long addr, size_t size)
393 {
394 	int diag;
395 	__u8  w1;
396 	__u16 w2;
397 	__u32 w4;
398 	__u64 w8;
399 	*word = 0;	/* Default value if addr or size is invalid */
400 
401 	switch (size) {
402 	case 1:
403 		diag = kdb_getphys(&w1, addr, sizeof(w1));
404 		if (!diag)
405 			*word = w1;
406 		break;
407 	case 2:
408 		diag = kdb_getphys(&w2, addr, sizeof(w2));
409 		if (!diag)
410 			*word = w2;
411 		break;
412 	case 4:
413 		diag = kdb_getphys(&w4, addr, sizeof(w4));
414 		if (!diag)
415 			*word = w4;
416 		break;
417 	case 8:
418 		if (size <= sizeof(*word)) {
419 			diag = kdb_getphys(&w8, addr, sizeof(w8));
420 			if (!diag)
421 				*word = w8;
422 			break;
423 		}
424 		fallthrough;
425 	default:
426 		diag = KDB_BADWIDTH;
427 		kdb_func_printf("bad width %zu\n", size);
428 	}
429 	return diag;
430 }
431 
432 /*
433  * kdb_getword - Read a binary value.  Unlike kdb_getarea, this treats
434  *	data as numbers.
435  * Inputs:
436  *	word	Pointer to the word to receive the result.
437  *	addr	Address of the area to copy.
438  *	size	Size of the area.
439  * Returns:
440  *	0 for success, < 0 for error.
441  */
442 int kdb_getword(unsigned long *word, unsigned long addr, size_t size)
443 {
444 	int diag;
445 	__u8  w1;
446 	__u16 w2;
447 	__u32 w4;
448 	__u64 w8;
449 	*word = 0;	/* Default value if addr or size is invalid */
450 	switch (size) {
451 	case 1:
452 		diag = kdb_getarea(w1, addr);
453 		if (!diag)
454 			*word = w1;
455 		break;
456 	case 2:
457 		diag = kdb_getarea(w2, addr);
458 		if (!diag)
459 			*word = w2;
460 		break;
461 	case 4:
462 		diag = kdb_getarea(w4, addr);
463 		if (!diag)
464 			*word = w4;
465 		break;
466 	case 8:
467 		if (size <= sizeof(*word)) {
468 			diag = kdb_getarea(w8, addr);
469 			if (!diag)
470 				*word = w8;
471 			break;
472 		}
473 		fallthrough;
474 	default:
475 		diag = KDB_BADWIDTH;
476 		kdb_func_printf("bad width %zu\n", size);
477 	}
478 	return diag;
479 }
480 
481 /*
482  * kdb_putword - Write a binary value.  Unlike kdb_putarea, this
483  *	treats data as numbers.
484  * Inputs:
485  *	addr	Address of the area to write to..
486  *	word	The value to set.
487  *	size	Size of the area.
488  * Returns:
489  *	0 for success, < 0 for error.
490  */
491 int kdb_putword(unsigned long addr, unsigned long word, size_t size)
492 {
493 	int diag;
494 	__u8  w1;
495 	__u16 w2;
496 	__u32 w4;
497 	__u64 w8;
498 	switch (size) {
499 	case 1:
500 		w1 = word;
501 		diag = kdb_putarea(addr, w1);
502 		break;
503 	case 2:
504 		w2 = word;
505 		diag = kdb_putarea(addr, w2);
506 		break;
507 	case 4:
508 		w4 = word;
509 		diag = kdb_putarea(addr, w4);
510 		break;
511 	case 8:
512 		if (size <= sizeof(word)) {
513 			w8 = word;
514 			diag = kdb_putarea(addr, w8);
515 			break;
516 		}
517 		fallthrough;
518 	default:
519 		diag = KDB_BADWIDTH;
520 		kdb_func_printf("bad width %zu\n", size);
521 	}
522 	return diag;
523 }
524 
525 /*
526  * kdb_task_state_string - Convert a string containing any of the
527  *	letters DRSTCZEUIMA to a mask for the process state field and
528  *	return the value.  If no argument is supplied, return the mask
529  *	that corresponds to environment variable PS, DRSTCZEU by
530  *	default.
531  * Inputs:
532  *	s	String to convert
533  * Returns:
534  *	Mask for process state.
535  * Notes:
536  *	The mask folds data from several sources into a single long value, so
537  *	be careful not to overlap the bits.  TASK_* bits are in the LSB,
538  *	special cases like UNRUNNABLE are in the MSB.  As of 2.6.10-rc1 there
539  *	is no overlap between TASK_* and EXIT_* but that may not always be
540  *	true, so EXIT_* bits are shifted left 16 bits before being stored in
541  *	the mask.
542  */
543 
544 /* unrunnable is < 0 */
545 #define UNRUNNABLE	(1UL << (8*sizeof(unsigned long) - 1))
546 #define RUNNING		(1UL << (8*sizeof(unsigned long) - 2))
547 #define IDLE		(1UL << (8*sizeof(unsigned long) - 3))
548 #define DAEMON		(1UL << (8*sizeof(unsigned long) - 4))
549 
550 unsigned long kdb_task_state_string(const char *s)
551 {
552 	long res = 0;
553 	if (!s) {
554 		s = kdbgetenv("PS");
555 		if (!s)
556 			s = "DRSTCZEU";	/* default value for ps */
557 	}
558 	while (*s) {
559 		switch (*s) {
560 		case 'D':
561 			res |= TASK_UNINTERRUPTIBLE;
562 			break;
563 		case 'R':
564 			res |= RUNNING;
565 			break;
566 		case 'S':
567 			res |= TASK_INTERRUPTIBLE;
568 			break;
569 		case 'T':
570 			res |= TASK_STOPPED;
571 			break;
572 		case 'C':
573 			res |= TASK_TRACED;
574 			break;
575 		case 'Z':
576 			res |= EXIT_ZOMBIE << 16;
577 			break;
578 		case 'E':
579 			res |= EXIT_DEAD << 16;
580 			break;
581 		case 'U':
582 			res |= UNRUNNABLE;
583 			break;
584 		case 'I':
585 			res |= IDLE;
586 			break;
587 		case 'M':
588 			res |= DAEMON;
589 			break;
590 		case 'A':
591 			res = ~0UL;
592 			break;
593 		default:
594 			  kdb_func_printf("unknown flag '%c' ignored\n", *s);
595 			  break;
596 		}
597 		++s;
598 	}
599 	return res;
600 }
601 
602 /*
603  * kdb_task_state_char - Return the character that represents the task state.
604  * Inputs:
605  *	p	struct task for the process
606  * Returns:
607  *	One character to represent the task state.
608  */
609 char kdb_task_state_char (const struct task_struct *p)
610 {
611 	unsigned int p_state;
612 	unsigned long tmp;
613 	char state;
614 	int cpu;
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 	p_state = READ_ONCE(p->__state);
622 	state = (p_state == 0) ? 'R' :
623 		(p_state < 0) ? 'U' :
624 		(p_state & TASK_UNINTERRUPTIBLE) ? 'D' :
625 		(p_state & TASK_STOPPED) ? 'T' :
626 		(p_state & TASK_TRACED) ? 'C' :
627 		(p->exit_state & EXIT_ZOMBIE) ? 'Z' :
628 		(p->exit_state & EXIT_DEAD) ? 'E' :
629 		(p_state & TASK_INTERRUPTIBLE) ? 'S' : '?';
630 	if (is_idle_task(p)) {
631 		/* Idle task.  Is it really idle, apart from the kdb
632 		 * interrupt? */
633 		if (!kdb_task_has_cpu(p) || kgdb_info[cpu].irq_depth == 1) {
634 			if (cpu != kdb_initial_cpu)
635 				state = 'I';	/* idle task */
636 		}
637 	} else if (!p->mm && state == 'S') {
638 		state = 'M';	/* sleeping system daemon */
639 	}
640 	return state;
641 }
642 
643 /*
644  * kdb_task_state - Return true if a process has the desired state
645  *	given by the mask.
646  * Inputs:
647  *	p	struct task for the process
648  *	mask	mask from kdb_task_state_string to select processes
649  * Returns:
650  *	True if the process matches at least one criteria defined by the mask.
651  */
652 unsigned long kdb_task_state(const struct task_struct *p, unsigned long mask)
653 {
654 	char state[] = { kdb_task_state_char(p), '\0' };
655 	return (mask & kdb_task_state_string(state)) != 0;
656 }
657 
658 /* Last ditch allocator for debugging, so we can still debug even when
659  * the GFP_ATOMIC pool has been exhausted.  The algorithms are tuned
660  * for space usage, not for speed.  One smallish memory pool, the free
661  * chain is always in ascending address order to allow coalescing,
662  * allocations are done in brute force best fit.
663  */
664 
665 struct debug_alloc_header {
666 	u32 next;	/* offset of next header from start of pool */
667 	u32 size;
668 	void *caller;
669 };
670 
671 /* The memory returned by this allocator must be aligned, which means
672  * so must the header size.  Do not assume that sizeof(struct
673  * debug_alloc_header) is a multiple of the alignment, explicitly
674  * calculate the overhead of this header, including the alignment.
675  * The rest of this code must not use sizeof() on any header or
676  * pointer to a header.
677  */
678 #define dah_align 8
679 #define dah_overhead ALIGN(sizeof(struct debug_alloc_header), dah_align)
680 
681 static u64 debug_alloc_pool_aligned[256*1024/dah_align];	/* 256K pool */
682 static char *debug_alloc_pool = (char *)debug_alloc_pool_aligned;
683 static u32 dah_first, dah_first_call = 1, dah_used, dah_used_max;
684 
685 /* Locking is awkward.  The debug code is called from all contexts,
686  * including non maskable interrupts.  A normal spinlock is not safe
687  * in NMI context.  Try to get the debug allocator lock, if it cannot
688  * be obtained after a second then give up.  If the lock could not be
689  * previously obtained on this cpu then only try once.
690  *
691  * sparse has no annotation for "this function _sometimes_ acquires a
692  * lock", so fudge the acquire/release notation.
693  */
694 static DEFINE_SPINLOCK(dap_lock);
695 static int get_dap_lock(void)
696 	__acquires(dap_lock)
697 {
698 	static int dap_locked = -1;
699 	int count;
700 	if (dap_locked == smp_processor_id())
701 		count = 1;
702 	else
703 		count = 1000;
704 	while (1) {
705 		if (spin_trylock(&dap_lock)) {
706 			dap_locked = -1;
707 			return 1;
708 		}
709 		if (!count--)
710 			break;
711 		udelay(1000);
712 	}
713 	dap_locked = smp_processor_id();
714 	__acquire(dap_lock);
715 	return 0;
716 }
717 
718 void *debug_kmalloc(size_t size, gfp_t flags)
719 {
720 	unsigned int rem, h_offset;
721 	struct debug_alloc_header *best, *bestprev, *prev, *h;
722 	void *p = NULL;
723 	if (!get_dap_lock()) {
724 		__release(dap_lock);	/* we never actually got it */
725 		return NULL;
726 	}
727 	h = (struct debug_alloc_header *)(debug_alloc_pool + dah_first);
728 	if (dah_first_call) {
729 		h->size = sizeof(debug_alloc_pool_aligned) - dah_overhead;
730 		dah_first_call = 0;
731 	}
732 	size = ALIGN(size, dah_align);
733 	prev = best = bestprev = NULL;
734 	while (1) {
735 		if (h->size >= size && (!best || h->size < best->size)) {
736 			best = h;
737 			bestprev = prev;
738 			if (h->size == size)
739 				break;
740 		}
741 		if (!h->next)
742 			break;
743 		prev = h;
744 		h = (struct debug_alloc_header *)(debug_alloc_pool + h->next);
745 	}
746 	if (!best)
747 		goto out;
748 	rem = best->size - size;
749 	/* The pool must always contain at least one header */
750 	if (best->next == 0 && bestprev == NULL && rem < dah_overhead)
751 		goto out;
752 	if (rem >= dah_overhead) {
753 		best->size = size;
754 		h_offset = ((char *)best - debug_alloc_pool) +
755 			   dah_overhead + best->size;
756 		h = (struct debug_alloc_header *)(debug_alloc_pool + h_offset);
757 		h->size = rem - dah_overhead;
758 		h->next = best->next;
759 	} else
760 		h_offset = best->next;
761 	best->caller = __builtin_return_address(0);
762 	dah_used += best->size;
763 	dah_used_max = max(dah_used, dah_used_max);
764 	if (bestprev)
765 		bestprev->next = h_offset;
766 	else
767 		dah_first = h_offset;
768 	p = (char *)best + dah_overhead;
769 	memset(p, POISON_INUSE, best->size - 1);
770 	*((char *)p + best->size - 1) = POISON_END;
771 out:
772 	spin_unlock(&dap_lock);
773 	return p;
774 }
775 
776 void debug_kfree(void *p)
777 {
778 	struct debug_alloc_header *h;
779 	unsigned int h_offset;
780 	if (!p)
781 		return;
782 	if ((char *)p < debug_alloc_pool ||
783 	    (char *)p >= debug_alloc_pool + sizeof(debug_alloc_pool_aligned)) {
784 		kfree(p);
785 		return;
786 	}
787 	if (!get_dap_lock()) {
788 		__release(dap_lock);	/* we never actually got it */
789 		return;		/* memory leak, cannot be helped */
790 	}
791 	h = (struct debug_alloc_header *)((char *)p - dah_overhead);
792 	memset(p, POISON_FREE, h->size - 1);
793 	*((char *)p + h->size - 1) = POISON_END;
794 	h->caller = NULL;
795 	dah_used -= h->size;
796 	h_offset = (char *)h - debug_alloc_pool;
797 	if (h_offset < dah_first) {
798 		h->next = dah_first;
799 		dah_first = h_offset;
800 	} else {
801 		struct debug_alloc_header *prev;
802 		unsigned int prev_offset;
803 		prev = (struct debug_alloc_header *)(debug_alloc_pool +
804 						     dah_first);
805 		while (1) {
806 			if (!prev->next || prev->next > h_offset)
807 				break;
808 			prev = (struct debug_alloc_header *)
809 				(debug_alloc_pool + prev->next);
810 		}
811 		prev_offset = (char *)prev - debug_alloc_pool;
812 		if (prev_offset + dah_overhead + prev->size == h_offset) {
813 			prev->size += dah_overhead + h->size;
814 			memset(h, POISON_FREE, dah_overhead - 1);
815 			*((char *)h + dah_overhead - 1) = POISON_END;
816 			h = prev;
817 			h_offset = prev_offset;
818 		} else {
819 			h->next = prev->next;
820 			prev->next = h_offset;
821 		}
822 	}
823 	if (h_offset + dah_overhead + h->size == h->next) {
824 		struct debug_alloc_header *next;
825 		next = (struct debug_alloc_header *)
826 			(debug_alloc_pool + h->next);
827 		h->size += dah_overhead + next->size;
828 		h->next = next->next;
829 		memset(next, POISON_FREE, dah_overhead - 1);
830 		*((char *)next + dah_overhead - 1) = POISON_END;
831 	}
832 	spin_unlock(&dap_lock);
833 }
834 
835 void debug_kusage(void)
836 {
837 	struct debug_alloc_header *h_free, *h_used;
838 #ifdef	CONFIG_IA64
839 	/* FIXME: using dah for ia64 unwind always results in a memory leak.
840 	 * Fix that memory leak first, then set debug_kusage_one_time = 1 for
841 	 * all architectures.
842 	 */
843 	static int debug_kusage_one_time;
844 #else
845 	static int debug_kusage_one_time = 1;
846 #endif
847 	if (!get_dap_lock()) {
848 		__release(dap_lock);	/* we never actually got it */
849 		return;
850 	}
851 	h_free = (struct debug_alloc_header *)(debug_alloc_pool + dah_first);
852 	if (dah_first == 0 &&
853 	    (h_free->size == sizeof(debug_alloc_pool_aligned) - dah_overhead ||
854 	     dah_first_call))
855 		goto out;
856 	if (!debug_kusage_one_time)
857 		goto out;
858 	debug_kusage_one_time = 0;
859 	kdb_func_printf("debug_kmalloc memory leak dah_first %d\n", dah_first);
860 	if (dah_first) {
861 		h_used = (struct debug_alloc_header *)debug_alloc_pool;
862 		kdb_func_printf("h_used %px size %d\n", h_used, h_used->size);
863 	}
864 	do {
865 		h_used = (struct debug_alloc_header *)
866 			  ((char *)h_free + dah_overhead + h_free->size);
867 		kdb_func_printf("h_used %px size %d caller %px\n",
868 				h_used, h_used->size, h_used->caller);
869 		h_free = (struct debug_alloc_header *)
870 			  (debug_alloc_pool + h_free->next);
871 	} while (h_free->next);
872 	h_used = (struct debug_alloc_header *)
873 		  ((char *)h_free + dah_overhead + h_free->size);
874 	if ((char *)h_used - debug_alloc_pool !=
875 	    sizeof(debug_alloc_pool_aligned))
876 		kdb_func_printf("h_used %px size %d caller %px\n",
877 				h_used, h_used->size, h_used->caller);
878 out:
879 	spin_unlock(&dap_lock);
880 }
881 
882 /* Maintain a small stack of kdb_flags to allow recursion without disturbing
883  * the global kdb state.
884  */
885 
886 static int kdb_flags_stack[4], kdb_flags_index;
887 
888 void kdb_save_flags(void)
889 {
890 	BUG_ON(kdb_flags_index >= ARRAY_SIZE(kdb_flags_stack));
891 	kdb_flags_stack[kdb_flags_index++] = kdb_flags;
892 }
893 
894 void kdb_restore_flags(void)
895 {
896 	BUG_ON(kdb_flags_index <= 0);
897 	kdb_flags = kdb_flags_stack[--kdb_flags_index];
898 }
899