xref: /openbmc/linux/arch/arm/mm/fault.c (revision fd589a8f)
1 /*
2  *  linux/arch/arm/mm/fault.c
3  *
4  *  Copyright (C) 1995  Linus Torvalds
5  *  Modifications for ARM processor (c) 1995-2004 Russell King
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/module.h>
12 #include <linux/signal.h>
13 #include <linux/mm.h>
14 #include <linux/hardirq.h>
15 #include <linux/init.h>
16 #include <linux/kprobes.h>
17 #include <linux/uaccess.h>
18 #include <linux/page-flags.h>
19 #include <linux/sched.h>
20 #include <linux/highmem.h>
21 
22 #include <asm/system.h>
23 #include <asm/pgtable.h>
24 #include <asm/tlbflush.h>
25 
26 #include "fault.h"
27 
28 #ifdef CONFIG_MMU
29 
30 #ifdef CONFIG_KPROBES
31 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
32 {
33 	int ret = 0;
34 
35 	if (!user_mode(regs)) {
36 		/* kprobe_running() needs smp_processor_id() */
37 		preempt_disable();
38 		if (kprobe_running() && kprobe_fault_handler(regs, fsr))
39 			ret = 1;
40 		preempt_enable();
41 	}
42 
43 	return ret;
44 }
45 #else
46 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
47 {
48 	return 0;
49 }
50 #endif
51 
52 /*
53  * This is useful to dump out the page tables associated with
54  * 'addr' in mm 'mm'.
55  */
56 void show_pte(struct mm_struct *mm, unsigned long addr)
57 {
58 	pgd_t *pgd;
59 
60 	if (!mm)
61 		mm = &init_mm;
62 
63 	printk(KERN_ALERT "pgd = %p\n", mm->pgd);
64 	pgd = pgd_offset(mm, addr);
65 	printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
66 
67 	do {
68 		pmd_t *pmd;
69 		pte_t *pte;
70 
71 		if (pgd_none(*pgd))
72 			break;
73 
74 		if (pgd_bad(*pgd)) {
75 			printk("(bad)");
76 			break;
77 		}
78 
79 		pmd = pmd_offset(pgd, addr);
80 		if (PTRS_PER_PMD != 1)
81 			printk(", *pmd=%08lx", pmd_val(*pmd));
82 
83 		if (pmd_none(*pmd))
84 			break;
85 
86 		if (pmd_bad(*pmd)) {
87 			printk("(bad)");
88 			break;
89 		}
90 
91 		/* We must not map this if we have highmem enabled */
92 		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
93 			break;
94 
95 		pte = pte_offset_map(pmd, addr);
96 		printk(", *pte=%08lx", pte_val(*pte));
97 		printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
98 		pte_unmap(pte);
99 	} while(0);
100 
101 	printk("\n");
102 }
103 #else					/* CONFIG_MMU */
104 void show_pte(struct mm_struct *mm, unsigned long addr)
105 { }
106 #endif					/* CONFIG_MMU */
107 
108 /*
109  * Oops.  The kernel tried to access some page that wasn't present.
110  */
111 static void
112 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
113 		  struct pt_regs *regs)
114 {
115 	/*
116 	 * Are we prepared to handle this kernel fault?
117 	 */
118 	if (fixup_exception(regs))
119 		return;
120 
121 	/*
122 	 * No handler, we'll have to terminate things with extreme prejudice.
123 	 */
124 	bust_spinlocks(1);
125 	printk(KERN_ALERT
126 		"Unable to handle kernel %s at virtual address %08lx\n",
127 		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
128 		"paging request", addr);
129 
130 	show_pte(mm, addr);
131 	die("Oops", regs, fsr);
132 	bust_spinlocks(0);
133 	do_exit(SIGKILL);
134 }
135 
136 /*
137  * Something tried to access memory that isn't in our memory map..
138  * User mode accesses just cause a SIGSEGV
139  */
140 static void
141 __do_user_fault(struct task_struct *tsk, unsigned long addr,
142 		unsigned int fsr, unsigned int sig, int code,
143 		struct pt_regs *regs)
144 {
145 	struct siginfo si;
146 
147 #ifdef CONFIG_DEBUG_USER
148 	if (user_debug & UDBG_SEGV) {
149 		printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
150 		       tsk->comm, sig, addr, fsr);
151 		show_pte(tsk->mm, addr);
152 		show_regs(regs);
153 	}
154 #endif
155 
156 	tsk->thread.address = addr;
157 	tsk->thread.error_code = fsr;
158 	tsk->thread.trap_no = 14;
159 	si.si_signo = sig;
160 	si.si_errno = 0;
161 	si.si_code = code;
162 	si.si_addr = (void __user *)addr;
163 	force_sig_info(sig, &si, tsk);
164 }
165 
166 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
167 {
168 	struct task_struct *tsk = current;
169 	struct mm_struct *mm = tsk->active_mm;
170 
171 	/*
172 	 * If we are in kernel mode at this point, we
173 	 * have no context to handle this fault with.
174 	 */
175 	if (user_mode(regs))
176 		__do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
177 	else
178 		__do_kernel_fault(mm, addr, fsr, regs);
179 }
180 
181 #ifdef CONFIG_MMU
182 #define VM_FAULT_BADMAP		0x010000
183 #define VM_FAULT_BADACCESS	0x020000
184 
185 static int
186 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
187 		struct task_struct *tsk)
188 {
189 	struct vm_area_struct *vma;
190 	int fault, mask;
191 
192 	vma = find_vma(mm, addr);
193 	fault = VM_FAULT_BADMAP;
194 	if (!vma)
195 		goto out;
196 	if (vma->vm_start > addr)
197 		goto check_stack;
198 
199 	/*
200 	 * Ok, we have a good vm_area for this
201 	 * memory access, so we can handle it.
202 	 */
203 good_area:
204 	if (fsr & (1 << 11)) /* write? */
205 		mask = VM_WRITE;
206 	else
207 		mask = VM_READ|VM_EXEC|VM_WRITE;
208 
209 	fault = VM_FAULT_BADACCESS;
210 	if (!(vma->vm_flags & mask))
211 		goto out;
212 
213 	/*
214 	 * If for any reason at all we couldn't handle
215 	 * the fault, make sure we exit gracefully rather
216 	 * than endlessly redo the fault.
217 	 */
218 survive:
219 	fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & (1 << 11)) ? FAULT_FLAG_WRITE : 0);
220 	if (unlikely(fault & VM_FAULT_ERROR)) {
221 		if (fault & VM_FAULT_OOM)
222 			goto out_of_memory;
223 		else if (fault & VM_FAULT_SIGBUS)
224 			return fault;
225 		BUG();
226 	}
227 	if (fault & VM_FAULT_MAJOR)
228 		tsk->maj_flt++;
229 	else
230 		tsk->min_flt++;
231 	return fault;
232 
233 out_of_memory:
234 	if (!is_global_init(tsk))
235 		goto out;
236 
237 	/*
238 	 * If we are out of memory for pid1, sleep for a while and retry
239 	 */
240 	up_read(&mm->mmap_sem);
241 	yield();
242 	down_read(&mm->mmap_sem);
243 	goto survive;
244 
245 check_stack:
246 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
247 		goto good_area;
248 out:
249 	return fault;
250 }
251 
252 static int __kprobes
253 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
254 {
255 	struct task_struct *tsk;
256 	struct mm_struct *mm;
257 	int fault, sig, code;
258 
259 	if (notify_page_fault(regs, fsr))
260 		return 0;
261 
262 	tsk = current;
263 	mm  = tsk->mm;
264 
265 	/*
266 	 * If we're in an interrupt or have no user
267 	 * context, we must not take the fault..
268 	 */
269 	if (in_atomic() || !mm)
270 		goto no_context;
271 
272 	/*
273 	 * As per x86, we may deadlock here.  However, since the kernel only
274 	 * validly references user space from well defined areas of the code,
275 	 * we can bug out early if this is from code which shouldn't.
276 	 */
277 	if (!down_read_trylock(&mm->mmap_sem)) {
278 		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
279 			goto no_context;
280 		down_read(&mm->mmap_sem);
281 	}
282 
283 	fault = __do_page_fault(mm, addr, fsr, tsk);
284 	up_read(&mm->mmap_sem);
285 
286 	/*
287 	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
288 	 */
289 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
290 		return 0;
291 
292 	/*
293 	 * If we are in kernel mode at this point, we
294 	 * have no context to handle this fault with.
295 	 */
296 	if (!user_mode(regs))
297 		goto no_context;
298 
299 	if (fault & VM_FAULT_OOM) {
300 		/*
301 		 * We ran out of memory, or some other thing
302 		 * happened to us that made us unable to handle
303 		 * the page fault gracefully.
304 		 */
305 		printk("VM: killing process %s\n", tsk->comm);
306 		do_group_exit(SIGKILL);
307 		return 0;
308 	}
309 	if (fault & VM_FAULT_SIGBUS) {
310 		/*
311 		 * We had some memory, but were unable to
312 		 * successfully fix up this page fault.
313 		 */
314 		sig = SIGBUS;
315 		code = BUS_ADRERR;
316 	} else {
317 		/*
318 		 * Something tried to access memory that
319 		 * isn't in our memory map..
320 		 */
321 		sig = SIGSEGV;
322 		code = fault == VM_FAULT_BADACCESS ?
323 			SEGV_ACCERR : SEGV_MAPERR;
324 	}
325 
326 	__do_user_fault(tsk, addr, fsr, sig, code, regs);
327 	return 0;
328 
329 no_context:
330 	__do_kernel_fault(mm, addr, fsr, regs);
331 	return 0;
332 }
333 #else					/* CONFIG_MMU */
334 static int
335 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
336 {
337 	return 0;
338 }
339 #endif					/* CONFIG_MMU */
340 
341 /*
342  * First Level Translation Fault Handler
343  *
344  * We enter here because the first level page table doesn't contain
345  * a valid entry for the address.
346  *
347  * If the address is in kernel space (>= TASK_SIZE), then we are
348  * probably faulting in the vmalloc() area.
349  *
350  * If the init_task's first level page tables contains the relevant
351  * entry, we copy the it to this task.  If not, we send the process
352  * a signal, fixup the exception, or oops the kernel.
353  *
354  * NOTE! We MUST NOT take any locks for this case. We may be in an
355  * interrupt or a critical region, and should only copy the information
356  * from the master page table, nothing more.
357  */
358 #ifdef CONFIG_MMU
359 static int __kprobes
360 do_translation_fault(unsigned long addr, unsigned int fsr,
361 		     struct pt_regs *regs)
362 {
363 	unsigned int index;
364 	pgd_t *pgd, *pgd_k;
365 	pmd_t *pmd, *pmd_k;
366 
367 	if (addr < TASK_SIZE)
368 		return do_page_fault(addr, fsr, regs);
369 
370 	index = pgd_index(addr);
371 
372 	/*
373 	 * FIXME: CP15 C1 is write only on ARMv3 architectures.
374 	 */
375 	pgd = cpu_get_pgd() + index;
376 	pgd_k = init_mm.pgd + index;
377 
378 	if (pgd_none(*pgd_k))
379 		goto bad_area;
380 
381 	if (!pgd_present(*pgd))
382 		set_pgd(pgd, *pgd_k);
383 
384 	pmd_k = pmd_offset(pgd_k, addr);
385 	pmd   = pmd_offset(pgd, addr);
386 
387 	if (pmd_none(*pmd_k))
388 		goto bad_area;
389 
390 	copy_pmd(pmd, pmd_k);
391 	return 0;
392 
393 bad_area:
394 	do_bad_area(addr, fsr, regs);
395 	return 0;
396 }
397 #else					/* CONFIG_MMU */
398 static int
399 do_translation_fault(unsigned long addr, unsigned int fsr,
400 		     struct pt_regs *regs)
401 {
402 	return 0;
403 }
404 #endif					/* CONFIG_MMU */
405 
406 /*
407  * Some section permission faults need to be handled gracefully.
408  * They can happen due to a __{get,put}_user during an oops.
409  */
410 static int
411 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
412 {
413 	do_bad_area(addr, fsr, regs);
414 	return 0;
415 }
416 
417 /*
418  * This abort handler always returns "fault".
419  */
420 static int
421 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
422 {
423 	return 1;
424 }
425 
426 static struct fsr_info {
427 	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
428 	int	sig;
429 	int	code;
430 	const char *name;
431 } fsr_info[] = {
432 	/*
433 	 * The following are the standard ARMv3 and ARMv4 aborts.  ARMv5
434 	 * defines these to be "precise" aborts.
435 	 */
436 	{ do_bad,		SIGSEGV, 0,		"vector exception"		   },
437 	{ do_bad,		SIGILL,	 BUS_ADRALN,	"alignment exception"		   },
438 	{ do_bad,		SIGKILL, 0,		"terminal exception"		   },
439 	{ do_bad,		SIGILL,	 BUS_ADRALN,	"alignment exception"		   },
440 	{ do_bad,		SIGBUS,	 0,		"external abort on linefetch"	   },
441 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"section translation fault"	   },
442 	{ do_bad,		SIGBUS,	 0,		"external abort on linefetch"	   },
443 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"page translation fault"	   },
444 	{ do_bad,		SIGBUS,	 0,		"external abort on non-linefetch"  },
445 	{ do_bad,		SIGSEGV, SEGV_ACCERR,	"section domain fault"		   },
446 	{ do_bad,		SIGBUS,	 0,		"external abort on non-linefetch"  },
447 	{ do_bad,		SIGSEGV, SEGV_ACCERR,	"page domain fault"		   },
448 	{ do_bad,		SIGBUS,	 0,		"external abort on translation"	   },
449 	{ do_sect_fault,	SIGSEGV, SEGV_ACCERR,	"section permission fault"	   },
450 	{ do_bad,		SIGBUS,	 0,		"external abort on translation"	   },
451 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"page permission fault"		   },
452 	/*
453 	 * The following are "imprecise" aborts, which are signalled by bit
454 	 * 10 of the FSR, and may not be recoverable.  These are only
455 	 * supported if the CPU abort handler supports bit 10.
456 	 */
457 	{ do_bad,		SIGBUS,  0,		"unknown 16"			   },
458 	{ do_bad,		SIGBUS,  0,		"unknown 17"			   },
459 	{ do_bad,		SIGBUS,  0,		"unknown 18"			   },
460 	{ do_bad,		SIGBUS,  0,		"unknown 19"			   },
461 	{ do_bad,		SIGBUS,  0,		"lock abort"			   }, /* xscale */
462 	{ do_bad,		SIGBUS,  0,		"unknown 21"			   },
463 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"imprecise external abort"	   }, /* xscale */
464 	{ do_bad,		SIGBUS,  0,		"unknown 23"			   },
465 	{ do_bad,		SIGBUS,  0,		"dcache parity error"		   }, /* xscale */
466 	{ do_bad,		SIGBUS,  0,		"unknown 25"			   },
467 	{ do_bad,		SIGBUS,  0,		"unknown 26"			   },
468 	{ do_bad,		SIGBUS,  0,		"unknown 27"			   },
469 	{ do_bad,		SIGBUS,  0,		"unknown 28"			   },
470 	{ do_bad,		SIGBUS,  0,		"unknown 29"			   },
471 	{ do_bad,		SIGBUS,  0,		"unknown 30"			   },
472 	{ do_bad,		SIGBUS,  0,		"unknown 31"			   }
473 };
474 
475 void __init
476 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
477 		int sig, const char *name)
478 {
479 	if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
480 		fsr_info[nr].fn   = fn;
481 		fsr_info[nr].sig  = sig;
482 		fsr_info[nr].name = name;
483 	}
484 }
485 
486 /*
487  * Dispatch a data abort to the relevant handler.
488  */
489 asmlinkage void __exception
490 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
491 {
492 	const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
493 	struct siginfo info;
494 
495 	if (!inf->fn(addr, fsr, regs))
496 		return;
497 
498 	printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
499 		inf->name, fsr, addr);
500 
501 	info.si_signo = inf->sig;
502 	info.si_errno = 0;
503 	info.si_code  = inf->code;
504 	info.si_addr  = (void __user *)addr;
505 	arm_notify_die("", regs, &info, fsr, 0);
506 }
507 
508 asmlinkage void __exception
509 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
510 {
511 	do_translation_fault(addr, 0, regs);
512 }
513 
514