xref: /openbmc/linux/kernel/debug/kdb/kdb_support.c (revision d4fd6347)
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 	if (KDB_DEBUG(AR))
43 		kdb_printf("kdbgetsymval: symname=%s, symtab=%px\n", symname,
44 			   symtab);
45 	memset(symtab, 0, sizeof(*symtab));
46 	symtab->sym_start = kallsyms_lookup_name(symname);
47 	if (symtab->sym_start) {
48 		if (KDB_DEBUG(AR))
49 			kdb_printf("kdbgetsymval: returns 1, "
50 				   "symtab->sym_start=0x%lx\n",
51 				   symtab->sym_start);
52 		return 1;
53 	}
54 	if (KDB_DEBUG(AR))
55 		kdb_printf("kdbgetsymval: returns 0\n");
56 	return 0;
57 }
58 EXPORT_SYMBOL(kdbgetsymval);
59 
60 static char *kdb_name_table[100];	/* arbitrary size */
61 
62 /*
63  * kdbnearsym -	Return the name of the symbol with the nearest address
64  *	less than 'addr'.
65  *
66  * Parameters:
67  *	addr	Address to check for symbol near
68  *	symtab  Structure to receive results
69  * Returns:
70  *	0	No sections contain this address, symtab zero filled
71  *	1	Address mapped to module/symbol/section, data in symtab
72  * Remarks:
73  *	2.6 kallsyms has a "feature" where it unpacks the name into a
74  *	string.  If that string is reused before the caller expects it
75  *	then the caller sees its string change without warning.  To
76  *	avoid cluttering up the main kdb code with lots of kdb_strdup,
77  *	tests and kfree calls, kdbnearsym maintains an LRU list of the
78  *	last few unique strings.  The list is sized large enough to
79  *	hold active strings, no kdb caller of kdbnearsym makes more
80  *	than ~20 later calls before using a saved value.
81  */
82 int kdbnearsym(unsigned long addr, kdb_symtab_t *symtab)
83 {
84 	int ret = 0;
85 	unsigned long symbolsize = 0;
86 	unsigned long offset = 0;
87 #define knt1_size 128		/* must be >= kallsyms table size */
88 	char *knt1 = NULL;
89 
90 	if (KDB_DEBUG(AR))
91 		kdb_printf("kdbnearsym: addr=0x%lx, symtab=%px\n", addr, symtab);
92 	memset(symtab, 0, sizeof(*symtab));
93 
94 	if (addr < 4096)
95 		goto out;
96 	knt1 = debug_kmalloc(knt1_size, GFP_ATOMIC);
97 	if (!knt1) {
98 		kdb_printf("kdbnearsym: addr=0x%lx cannot kmalloc knt1\n",
99 			   addr);
100 		goto out;
101 	}
102 	symtab->sym_name = kallsyms_lookup(addr, &symbolsize , &offset,
103 				(char **)(&symtab->mod_name), knt1);
104 	if (offset > 8*1024*1024) {
105 		symtab->sym_name = NULL;
106 		addr = offset = symbolsize = 0;
107 	}
108 	symtab->sym_start = addr - offset;
109 	symtab->sym_end = symtab->sym_start + symbolsize;
110 	ret = symtab->sym_name != NULL && *(symtab->sym_name) != '\0';
111 
112 	if (ret) {
113 		int i;
114 		/* Another 2.6 kallsyms "feature".  Sometimes the sym_name is
115 		 * set but the buffer passed into kallsyms_lookup is not used,
116 		 * so it contains garbage.  The caller has to work out which
117 		 * buffer needs to be saved.
118 		 *
119 		 * What was Rusty smoking when he wrote that code?
120 		 */
121 		if (symtab->sym_name != knt1) {
122 			strncpy(knt1, symtab->sym_name, knt1_size);
123 			knt1[knt1_size-1] = '\0';
124 		}
125 		for (i = 0; i < ARRAY_SIZE(kdb_name_table); ++i) {
126 			if (kdb_name_table[i] &&
127 			    strcmp(kdb_name_table[i], knt1) == 0)
128 				break;
129 		}
130 		if (i >= ARRAY_SIZE(kdb_name_table)) {
131 			debug_kfree(kdb_name_table[0]);
132 			memmove(kdb_name_table, kdb_name_table+1,
133 			       sizeof(kdb_name_table[0]) *
134 			       (ARRAY_SIZE(kdb_name_table)-1));
135 		} else {
136 			debug_kfree(knt1);
137 			knt1 = kdb_name_table[i];
138 			memmove(kdb_name_table+i, kdb_name_table+i+1,
139 			       sizeof(kdb_name_table[0]) *
140 			       (ARRAY_SIZE(kdb_name_table)-i-1));
141 		}
142 		i = ARRAY_SIZE(kdb_name_table) - 1;
143 		kdb_name_table[i] = knt1;
144 		symtab->sym_name = kdb_name_table[i];
145 		knt1 = NULL;
146 	}
147 
148 	if (symtab->mod_name == NULL)
149 		symtab->mod_name = "kernel";
150 	if (KDB_DEBUG(AR))
151 		kdb_printf("kdbnearsym: returns %d symtab->sym_start=0x%lx, "
152 		   "symtab->mod_name=%px, symtab->sym_name=%px (%s)\n", ret,
153 		   symtab->sym_start, symtab->mod_name, symtab->sym_name,
154 		   symtab->sym_name);
155 
156 out:
157 	debug_kfree(knt1);
158 	return ret;
159 }
160 
161 void kdbnearsym_cleanup(void)
162 {
163 	int i;
164 	for (i = 0; i < ARRAY_SIZE(kdb_name_table); ++i) {
165 		if (kdb_name_table[i]) {
166 			debug_kfree(kdb_name_table[i]);
167 			kdb_name_table[i] = NULL;
168 		}
169 	}
170 }
171 
172 static char ks_namebuf[KSYM_NAME_LEN+1], ks_namebuf_prev[KSYM_NAME_LEN+1];
173 
174 /*
175  * kallsyms_symbol_complete
176  *
177  * Parameters:
178  *	prefix_name	prefix of a symbol name to lookup
179  *	max_len		maximum length that can be returned
180  * Returns:
181  *	Number of symbols which match the given prefix.
182  * Notes:
183  *	prefix_name is changed to contain the longest unique prefix that
184  *	starts with this prefix (tab completion).
185  */
186 int kallsyms_symbol_complete(char *prefix_name, int max_len)
187 {
188 	loff_t pos = 0;
189 	int prefix_len = strlen(prefix_name), prev_len = 0;
190 	int i, number = 0;
191 	const char *name;
192 
193 	while ((name = kdb_walk_kallsyms(&pos))) {
194 		if (strncmp(name, prefix_name, prefix_len) == 0) {
195 			strscpy(ks_namebuf, name, sizeof(ks_namebuf));
196 			/* Work out the longest name that matches the prefix */
197 			if (++number == 1) {
198 				prev_len = min_t(int, max_len-1,
199 						 strlen(ks_namebuf));
200 				memcpy(ks_namebuf_prev, ks_namebuf, prev_len);
201 				ks_namebuf_prev[prev_len] = '\0';
202 				continue;
203 			}
204 			for (i = 0; i < prev_len; i++) {
205 				if (ks_namebuf[i] != ks_namebuf_prev[i]) {
206 					prev_len = i;
207 					ks_namebuf_prev[i] = '\0';
208 					break;
209 				}
210 			}
211 		}
212 	}
213 	if (prev_len > prefix_len)
214 		memcpy(prefix_name, ks_namebuf_prev, prev_len+1);
215 	return number;
216 }
217 
218 /*
219  * kallsyms_symbol_next
220  *
221  * Parameters:
222  *	prefix_name	prefix of a symbol name to lookup
223  *	flag	0 means search from the head, 1 means continue search.
224  *	buf_size	maximum length that can be written to prefix_name
225  *			buffer
226  * Returns:
227  *	1 if a symbol matches the given prefix.
228  *	0 if no string found
229  */
230 int kallsyms_symbol_next(char *prefix_name, int flag, int buf_size)
231 {
232 	int prefix_len = strlen(prefix_name);
233 	static loff_t pos;
234 	const char *name;
235 
236 	if (!flag)
237 		pos = 0;
238 
239 	while ((name = kdb_walk_kallsyms(&pos))) {
240 		if (!strncmp(name, prefix_name, prefix_len))
241 			return strscpy(prefix_name, name, buf_size);
242 	}
243 	return 0;
244 }
245 
246 /*
247  * kdb_symbol_print - Standard method for printing a symbol name and offset.
248  * Inputs:
249  *	addr	Address to be printed.
250  *	symtab	Address of symbol data, if NULL this routine does its
251  *		own lookup.
252  *	punc	Punctuation for string, bit field.
253  * Remarks:
254  *	The string and its punctuation is only printed if the address
255  *	is inside the kernel, except that the value is always printed
256  *	when requested.
257  */
258 void kdb_symbol_print(unsigned long addr, const kdb_symtab_t *symtab_p,
259 		      unsigned int punc)
260 {
261 	kdb_symtab_t symtab, *symtab_p2;
262 	if (symtab_p) {
263 		symtab_p2 = (kdb_symtab_t *)symtab_p;
264 	} else {
265 		symtab_p2 = &symtab;
266 		kdbnearsym(addr, symtab_p2);
267 	}
268 	if (!(symtab_p2->sym_name || (punc & KDB_SP_VALUE)))
269 		return;
270 	if (punc & KDB_SP_SPACEB)
271 		kdb_printf(" ");
272 	if (punc & KDB_SP_VALUE)
273 		kdb_printf(kdb_machreg_fmt0, addr);
274 	if (symtab_p2->sym_name) {
275 		if (punc & KDB_SP_VALUE)
276 			kdb_printf(" ");
277 		if (punc & KDB_SP_PAREN)
278 			kdb_printf("(");
279 		if (strcmp(symtab_p2->mod_name, "kernel"))
280 			kdb_printf("[%s]", symtab_p2->mod_name);
281 		kdb_printf("%s", symtab_p2->sym_name);
282 		if (addr != symtab_p2->sym_start)
283 			kdb_printf("+0x%lx", addr - symtab_p2->sym_start);
284 		if (punc & KDB_SP_SYMSIZE)
285 			kdb_printf("/0x%lx",
286 				   symtab_p2->sym_end - symtab_p2->sym_start);
287 		if (punc & KDB_SP_PAREN)
288 			kdb_printf(")");
289 	}
290 	if (punc & KDB_SP_SPACEA)
291 		kdb_printf(" ");
292 	if (punc & KDB_SP_NEWLINE)
293 		kdb_printf("\n");
294 }
295 
296 /*
297  * kdb_strdup - kdb equivalent of strdup, for disasm code.
298  * Inputs:
299  *	str	The string to duplicate.
300  *	type	Flags to kmalloc for the new string.
301  * Returns:
302  *	Address of the new string, NULL if storage could not be allocated.
303  * Remarks:
304  *	This is not in lib/string.c because it uses kmalloc which is not
305  *	available when string.o is used in boot loaders.
306  */
307 char *kdb_strdup(const char *str, gfp_t type)
308 {
309 	int n = strlen(str)+1;
310 	char *s = kmalloc(n, type);
311 	if (!s)
312 		return NULL;
313 	return strcpy(s, str);
314 }
315 
316 /*
317  * kdb_getarea_size - Read an area of data.  The kdb equivalent of
318  *	copy_from_user, with kdb messages for invalid addresses.
319  * Inputs:
320  *	res	Pointer to the area to receive the result.
321  *	addr	Address of the area to copy.
322  *	size	Size of the area.
323  * Returns:
324  *	0 for success, < 0 for error.
325  */
326 int kdb_getarea_size(void *res, unsigned long addr, size_t size)
327 {
328 	int ret = probe_kernel_read((char *)res, (char *)addr, size);
329 	if (ret) {
330 		if (!KDB_STATE(SUPPRESS)) {
331 			kdb_printf("kdb_getarea: Bad address 0x%lx\n", addr);
332 			KDB_STATE_SET(SUPPRESS);
333 		}
334 		ret = KDB_BADADDR;
335 	} else {
336 		KDB_STATE_CLEAR(SUPPRESS);
337 	}
338 	return ret;
339 }
340 
341 /*
342  * kdb_putarea_size - Write an area of data.  The kdb equivalent of
343  *	copy_to_user, with kdb messages for invalid addresses.
344  * Inputs:
345  *	addr	Address of the area to write to.
346  *	res	Pointer to the area holding the data.
347  *	size	Size of the area.
348  * Returns:
349  *	0 for success, < 0 for error.
350  */
351 int kdb_putarea_size(unsigned long addr, void *res, size_t size)
352 {
353 	int ret = probe_kernel_read((char *)addr, (char *)res, size);
354 	if (ret) {
355 		if (!KDB_STATE(SUPPRESS)) {
356 			kdb_printf("kdb_putarea: Bad address 0x%lx\n", addr);
357 			KDB_STATE_SET(SUPPRESS);
358 		}
359 		ret = KDB_BADADDR;
360 	} else {
361 		KDB_STATE_CLEAR(SUPPRESS);
362 	}
363 	return ret;
364 }
365 
366 /*
367  * kdb_getphys - Read data from a physical address. Validate the
368  * 	address is in range, use kmap_atomic() to get data
369  * 	similar to kdb_getarea() - but for phys addresses
370  * Inputs:
371  * 	res	Pointer to the word to receive the result
372  * 	addr	Physical address of the area to copy
373  * 	size	Size of the area
374  * Returns:
375  *	0 for success, < 0 for error.
376  */
377 static int kdb_getphys(void *res, unsigned long addr, size_t size)
378 {
379 	unsigned long pfn;
380 	void *vaddr;
381 	struct page *page;
382 
383 	pfn = (addr >> PAGE_SHIFT);
384 	if (!pfn_valid(pfn))
385 		return 1;
386 	page = pfn_to_page(pfn);
387 	vaddr = kmap_atomic(page);
388 	memcpy(res, vaddr + (addr & (PAGE_SIZE - 1)), size);
389 	kunmap_atomic(vaddr);
390 
391 	return 0;
392 }
393 
394 /*
395  * kdb_getphysword
396  * Inputs:
397  *	word	Pointer to the word to receive the result.
398  *	addr	Address of the area to copy.
399  *	size	Size of the area.
400  * Returns:
401  *	0 for success, < 0 for error.
402  */
403 int kdb_getphysword(unsigned long *word, unsigned long addr, size_t size)
404 {
405 	int diag;
406 	__u8  w1;
407 	__u16 w2;
408 	__u32 w4;
409 	__u64 w8;
410 	*word = 0;	/* Default value if addr or size is invalid */
411 
412 	switch (size) {
413 	case 1:
414 		diag = kdb_getphys(&w1, addr, sizeof(w1));
415 		if (!diag)
416 			*word = w1;
417 		break;
418 	case 2:
419 		diag = kdb_getphys(&w2, addr, sizeof(w2));
420 		if (!diag)
421 			*word = w2;
422 		break;
423 	case 4:
424 		diag = kdb_getphys(&w4, addr, sizeof(w4));
425 		if (!diag)
426 			*word = w4;
427 		break;
428 	case 8:
429 		if (size <= sizeof(*word)) {
430 			diag = kdb_getphys(&w8, addr, sizeof(w8));
431 			if (!diag)
432 				*word = w8;
433 			break;
434 		}
435 		/* fall through */
436 	default:
437 		diag = KDB_BADWIDTH;
438 		kdb_printf("kdb_getphysword: bad width %ld\n", (long) size);
439 	}
440 	return diag;
441 }
442 
443 /*
444  * kdb_getword - Read a binary value.  Unlike kdb_getarea, this treats
445  *	data as numbers.
446  * Inputs:
447  *	word	Pointer to the word to receive the result.
448  *	addr	Address of the area to copy.
449  *	size	Size of the area.
450  * Returns:
451  *	0 for success, < 0 for error.
452  */
453 int kdb_getword(unsigned long *word, unsigned long addr, size_t size)
454 {
455 	int diag;
456 	__u8  w1;
457 	__u16 w2;
458 	__u32 w4;
459 	__u64 w8;
460 	*word = 0;	/* Default value if addr or size is invalid */
461 	switch (size) {
462 	case 1:
463 		diag = kdb_getarea(w1, addr);
464 		if (!diag)
465 			*word = w1;
466 		break;
467 	case 2:
468 		diag = kdb_getarea(w2, addr);
469 		if (!diag)
470 			*word = w2;
471 		break;
472 	case 4:
473 		diag = kdb_getarea(w4, addr);
474 		if (!diag)
475 			*word = w4;
476 		break;
477 	case 8:
478 		if (size <= sizeof(*word)) {
479 			diag = kdb_getarea(w8, addr);
480 			if (!diag)
481 				*word = w8;
482 			break;
483 		}
484 		/* fall through */
485 	default:
486 		diag = KDB_BADWIDTH;
487 		kdb_printf("kdb_getword: bad width %ld\n", (long) size);
488 	}
489 	return diag;
490 }
491 
492 /*
493  * kdb_putword - Write a binary value.  Unlike kdb_putarea, this
494  *	treats data as numbers.
495  * Inputs:
496  *	addr	Address of the area to write to..
497  *	word	The value to set.
498  *	size	Size of the area.
499  * Returns:
500  *	0 for success, < 0 for error.
501  */
502 int kdb_putword(unsigned long addr, unsigned long word, size_t size)
503 {
504 	int diag;
505 	__u8  w1;
506 	__u16 w2;
507 	__u32 w4;
508 	__u64 w8;
509 	switch (size) {
510 	case 1:
511 		w1 = word;
512 		diag = kdb_putarea(addr, w1);
513 		break;
514 	case 2:
515 		w2 = word;
516 		diag = kdb_putarea(addr, w2);
517 		break;
518 	case 4:
519 		w4 = word;
520 		diag = kdb_putarea(addr, w4);
521 		break;
522 	case 8:
523 		if (size <= sizeof(word)) {
524 			w8 = word;
525 			diag = kdb_putarea(addr, w8);
526 			break;
527 		}
528 		/* fall through */
529 	default:
530 		diag = KDB_BADWIDTH;
531 		kdb_printf("kdb_putword: bad width %ld\n", (long) size);
532 	}
533 	return diag;
534 }
535 
536 /*
537  * kdb_task_state_string - Convert a string containing any of the
538  *	letters DRSTCZEUIMA to a mask for the process state field and
539  *	return the value.  If no argument is supplied, return the mask
540  *	that corresponds to environment variable PS, DRSTCZEU by
541  *	default.
542  * Inputs:
543  *	s	String to convert
544  * Returns:
545  *	Mask for process state.
546  * Notes:
547  *	The mask folds data from several sources into a single long value, so
548  *	be careful not to overlap the bits.  TASK_* bits are in the LSB,
549  *	special cases like UNRUNNABLE are in the MSB.  As of 2.6.10-rc1 there
550  *	is no overlap between TASK_* and EXIT_* but that may not always be
551  *	true, so EXIT_* bits are shifted left 16 bits before being stored in
552  *	the mask.
553  */
554 
555 /* unrunnable is < 0 */
556 #define UNRUNNABLE	(1UL << (8*sizeof(unsigned long) - 1))
557 #define RUNNING		(1UL << (8*sizeof(unsigned long) - 2))
558 #define IDLE		(1UL << (8*sizeof(unsigned long) - 3))
559 #define DAEMON		(1UL << (8*sizeof(unsigned long) - 4))
560 
561 unsigned long kdb_task_state_string(const char *s)
562 {
563 	long res = 0;
564 	if (!s) {
565 		s = kdbgetenv("PS");
566 		if (!s)
567 			s = "DRSTCZEU";	/* default value for ps */
568 	}
569 	while (*s) {
570 		switch (*s) {
571 		case 'D':
572 			res |= TASK_UNINTERRUPTIBLE;
573 			break;
574 		case 'R':
575 			res |= RUNNING;
576 			break;
577 		case 'S':
578 			res |= TASK_INTERRUPTIBLE;
579 			break;
580 		case 'T':
581 			res |= TASK_STOPPED;
582 			break;
583 		case 'C':
584 			res |= TASK_TRACED;
585 			break;
586 		case 'Z':
587 			res |= EXIT_ZOMBIE << 16;
588 			break;
589 		case 'E':
590 			res |= EXIT_DEAD << 16;
591 			break;
592 		case 'U':
593 			res |= UNRUNNABLE;
594 			break;
595 		case 'I':
596 			res |= IDLE;
597 			break;
598 		case 'M':
599 			res |= DAEMON;
600 			break;
601 		case 'A':
602 			res = ~0UL;
603 			break;
604 		default:
605 			  kdb_printf("%s: unknown flag '%c' ignored\n",
606 				     __func__, *s);
607 			  break;
608 		}
609 		++s;
610 	}
611 	return res;
612 }
613 
614 /*
615  * kdb_task_state_char - Return the character that represents the task state.
616  * Inputs:
617  *	p	struct task for the process
618  * Returns:
619  *	One character to represent the task state.
620  */
621 char kdb_task_state_char (const struct task_struct *p)
622 {
623 	int cpu;
624 	char state;
625 	unsigned long tmp;
626 
627 	if (!p || probe_kernel_read(&tmp, (char *)p, sizeof(unsigned long)))
628 		return 'E';
629 
630 	cpu = kdb_process_cpu(p);
631 	state = (p->state == 0) ? 'R' :
632 		(p->state < 0) ? 'U' :
633 		(p->state & TASK_UNINTERRUPTIBLE) ? 'D' :
634 		(p->state & TASK_STOPPED) ? 'T' :
635 		(p->state & TASK_TRACED) ? 'C' :
636 		(p->exit_state & EXIT_ZOMBIE) ? 'Z' :
637 		(p->exit_state & EXIT_DEAD) ? 'E' :
638 		(p->state & TASK_INTERRUPTIBLE) ? 'S' : '?';
639 	if (is_idle_task(p)) {
640 		/* Idle task.  Is it really idle, apart from the kdb
641 		 * interrupt? */
642 		if (!kdb_task_has_cpu(p) || kgdb_info[cpu].irq_depth == 1) {
643 			if (cpu != kdb_initial_cpu)
644 				state = 'I';	/* idle task */
645 		}
646 	} else if (!p->mm && state == 'S') {
647 		state = 'M';	/* sleeping system daemon */
648 	}
649 	return state;
650 }
651 
652 /*
653  * kdb_task_state - Return true if a process has the desired state
654  *	given by the mask.
655  * Inputs:
656  *	p	struct task for the process
657  *	mask	mask from kdb_task_state_string to select processes
658  * Returns:
659  *	True if the process matches at least one criteria defined by the mask.
660  */
661 unsigned long kdb_task_state(const struct task_struct *p, unsigned long mask)
662 {
663 	char state[] = { kdb_task_state_char(p), '\0' };
664 	return (mask & kdb_task_state_string(state)) != 0;
665 }
666 
667 /*
668  * kdb_print_nameval - Print a name and its value, converting the
669  *	value to a symbol lookup if possible.
670  * Inputs:
671  *	name	field name to print
672  *	val	value of field
673  */
674 void kdb_print_nameval(const char *name, unsigned long val)
675 {
676 	kdb_symtab_t symtab;
677 	kdb_printf("  %-11.11s ", name);
678 	if (kdbnearsym(val, &symtab))
679 		kdb_symbol_print(val, &symtab,
680 				 KDB_SP_VALUE|KDB_SP_SYMSIZE|KDB_SP_NEWLINE);
681 	else
682 		kdb_printf("0x%lx\n", val);
683 }
684 
685 /* Last ditch allocator for debugging, so we can still debug even when
686  * the GFP_ATOMIC pool has been exhausted.  The algorithms are tuned
687  * for space usage, not for speed.  One smallish memory pool, the free
688  * chain is always in ascending address order to allow coalescing,
689  * allocations are done in brute force best fit.
690  */
691 
692 struct debug_alloc_header {
693 	u32 next;	/* offset of next header from start of pool */
694 	u32 size;
695 	void *caller;
696 };
697 
698 /* The memory returned by this allocator must be aligned, which means
699  * so must the header size.  Do not assume that sizeof(struct
700  * debug_alloc_header) is a multiple of the alignment, explicitly
701  * calculate the overhead of this header, including the alignment.
702  * The rest of this code must not use sizeof() on any header or
703  * pointer to a header.
704  */
705 #define dah_align 8
706 #define dah_overhead ALIGN(sizeof(struct debug_alloc_header), dah_align)
707 
708 static u64 debug_alloc_pool_aligned[256*1024/dah_align];	/* 256K pool */
709 static char *debug_alloc_pool = (char *)debug_alloc_pool_aligned;
710 static u32 dah_first, dah_first_call = 1, dah_used, dah_used_max;
711 
712 /* Locking is awkward.  The debug code is called from all contexts,
713  * including non maskable interrupts.  A normal spinlock is not safe
714  * in NMI context.  Try to get the debug allocator lock, if it cannot
715  * be obtained after a second then give up.  If the lock could not be
716  * previously obtained on this cpu then only try once.
717  *
718  * sparse has no annotation for "this function _sometimes_ acquires a
719  * lock", so fudge the acquire/release notation.
720  */
721 static DEFINE_SPINLOCK(dap_lock);
722 static int get_dap_lock(void)
723 	__acquires(dap_lock)
724 {
725 	static int dap_locked = -1;
726 	int count;
727 	if (dap_locked == smp_processor_id())
728 		count = 1;
729 	else
730 		count = 1000;
731 	while (1) {
732 		if (spin_trylock(&dap_lock)) {
733 			dap_locked = -1;
734 			return 1;
735 		}
736 		if (!count--)
737 			break;
738 		udelay(1000);
739 	}
740 	dap_locked = smp_processor_id();
741 	__acquire(dap_lock);
742 	return 0;
743 }
744 
745 void *debug_kmalloc(size_t size, gfp_t flags)
746 {
747 	unsigned int rem, h_offset;
748 	struct debug_alloc_header *best, *bestprev, *prev, *h;
749 	void *p = NULL;
750 	if (!get_dap_lock()) {
751 		__release(dap_lock);	/* we never actually got it */
752 		return NULL;
753 	}
754 	h = (struct debug_alloc_header *)(debug_alloc_pool + dah_first);
755 	if (dah_first_call) {
756 		h->size = sizeof(debug_alloc_pool_aligned) - dah_overhead;
757 		dah_first_call = 0;
758 	}
759 	size = ALIGN(size, dah_align);
760 	prev = best = bestprev = NULL;
761 	while (1) {
762 		if (h->size >= size && (!best || h->size < best->size)) {
763 			best = h;
764 			bestprev = prev;
765 			if (h->size == size)
766 				break;
767 		}
768 		if (!h->next)
769 			break;
770 		prev = h;
771 		h = (struct debug_alloc_header *)(debug_alloc_pool + h->next);
772 	}
773 	if (!best)
774 		goto out;
775 	rem = best->size - size;
776 	/* The pool must always contain at least one header */
777 	if (best->next == 0 && bestprev == NULL && rem < dah_overhead)
778 		goto out;
779 	if (rem >= dah_overhead) {
780 		best->size = size;
781 		h_offset = ((char *)best - debug_alloc_pool) +
782 			   dah_overhead + best->size;
783 		h = (struct debug_alloc_header *)(debug_alloc_pool + h_offset);
784 		h->size = rem - dah_overhead;
785 		h->next = best->next;
786 	} else
787 		h_offset = best->next;
788 	best->caller = __builtin_return_address(0);
789 	dah_used += best->size;
790 	dah_used_max = max(dah_used, dah_used_max);
791 	if (bestprev)
792 		bestprev->next = h_offset;
793 	else
794 		dah_first = h_offset;
795 	p = (char *)best + dah_overhead;
796 	memset(p, POISON_INUSE, best->size - 1);
797 	*((char *)p + best->size - 1) = POISON_END;
798 out:
799 	spin_unlock(&dap_lock);
800 	return p;
801 }
802 
803 void debug_kfree(void *p)
804 {
805 	struct debug_alloc_header *h;
806 	unsigned int h_offset;
807 	if (!p)
808 		return;
809 	if ((char *)p < debug_alloc_pool ||
810 	    (char *)p >= debug_alloc_pool + sizeof(debug_alloc_pool_aligned)) {
811 		kfree(p);
812 		return;
813 	}
814 	if (!get_dap_lock()) {
815 		__release(dap_lock);	/* we never actually got it */
816 		return;		/* memory leak, cannot be helped */
817 	}
818 	h = (struct debug_alloc_header *)((char *)p - dah_overhead);
819 	memset(p, POISON_FREE, h->size - 1);
820 	*((char *)p + h->size - 1) = POISON_END;
821 	h->caller = NULL;
822 	dah_used -= h->size;
823 	h_offset = (char *)h - debug_alloc_pool;
824 	if (h_offset < dah_first) {
825 		h->next = dah_first;
826 		dah_first = h_offset;
827 	} else {
828 		struct debug_alloc_header *prev;
829 		unsigned int prev_offset;
830 		prev = (struct debug_alloc_header *)(debug_alloc_pool +
831 						     dah_first);
832 		while (1) {
833 			if (!prev->next || prev->next > h_offset)
834 				break;
835 			prev = (struct debug_alloc_header *)
836 				(debug_alloc_pool + prev->next);
837 		}
838 		prev_offset = (char *)prev - debug_alloc_pool;
839 		if (prev_offset + dah_overhead + prev->size == h_offset) {
840 			prev->size += dah_overhead + h->size;
841 			memset(h, POISON_FREE, dah_overhead - 1);
842 			*((char *)h + dah_overhead - 1) = POISON_END;
843 			h = prev;
844 			h_offset = prev_offset;
845 		} else {
846 			h->next = prev->next;
847 			prev->next = h_offset;
848 		}
849 	}
850 	if (h_offset + dah_overhead + h->size == h->next) {
851 		struct debug_alloc_header *next;
852 		next = (struct debug_alloc_header *)
853 			(debug_alloc_pool + h->next);
854 		h->size += dah_overhead + next->size;
855 		h->next = next->next;
856 		memset(next, POISON_FREE, dah_overhead - 1);
857 		*((char *)next + dah_overhead - 1) = POISON_END;
858 	}
859 	spin_unlock(&dap_lock);
860 }
861 
862 void debug_kusage(void)
863 {
864 	struct debug_alloc_header *h_free, *h_used;
865 #ifdef	CONFIG_IA64
866 	/* FIXME: using dah for ia64 unwind always results in a memory leak.
867 	 * Fix that memory leak first, then set debug_kusage_one_time = 1 for
868 	 * all architectures.
869 	 */
870 	static int debug_kusage_one_time;
871 #else
872 	static int debug_kusage_one_time = 1;
873 #endif
874 	if (!get_dap_lock()) {
875 		__release(dap_lock);	/* we never actually got it */
876 		return;
877 	}
878 	h_free = (struct debug_alloc_header *)(debug_alloc_pool + dah_first);
879 	if (dah_first == 0 &&
880 	    (h_free->size == sizeof(debug_alloc_pool_aligned) - dah_overhead ||
881 	     dah_first_call))
882 		goto out;
883 	if (!debug_kusage_one_time)
884 		goto out;
885 	debug_kusage_one_time = 0;
886 	kdb_printf("%s: debug_kmalloc memory leak dah_first %d\n",
887 		   __func__, dah_first);
888 	if (dah_first) {
889 		h_used = (struct debug_alloc_header *)debug_alloc_pool;
890 		kdb_printf("%s: h_used %px size %d\n", __func__, h_used,
891 			   h_used->size);
892 	}
893 	do {
894 		h_used = (struct debug_alloc_header *)
895 			  ((char *)h_free + dah_overhead + h_free->size);
896 		kdb_printf("%s: h_used %px size %d caller %px\n",
897 			   __func__, h_used, h_used->size, h_used->caller);
898 		h_free = (struct debug_alloc_header *)
899 			  (debug_alloc_pool + h_free->next);
900 	} while (h_free->next);
901 	h_used = (struct debug_alloc_header *)
902 		  ((char *)h_free + dah_overhead + h_free->size);
903 	if ((char *)h_used - debug_alloc_pool !=
904 	    sizeof(debug_alloc_pool_aligned))
905 		kdb_printf("%s: h_used %px size %d caller %px\n",
906 			   __func__, h_used, h_used->size, h_used->caller);
907 out:
908 	spin_unlock(&dap_lock);
909 }
910 
911 /* Maintain a small stack of kdb_flags to allow recursion without disturbing
912  * the global kdb state.
913  */
914 
915 static int kdb_flags_stack[4], kdb_flags_index;
916 
917 void kdb_save_flags(void)
918 {
919 	BUG_ON(kdb_flags_index >= ARRAY_SIZE(kdb_flags_stack));
920 	kdb_flags_stack[kdb_flags_index++] = kdb_flags;
921 }
922 
923 void kdb_restore_flags(void)
924 {
925 	BUG_ON(kdb_flags_index <= 0);
926 	kdb_flags = kdb_flags_stack[--kdb_flags_index];
927 }
928