xref: /openbmc/linux/arch/arm/mm/fault.c (revision 67306da6)
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 
19 #include <asm/system.h>
20 #include <asm/pgtable.h>
21 #include <asm/tlbflush.h>
22 
23 #include "fault.h"
24 
25 
26 #ifdef CONFIG_KPROBES
27 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
28 {
29 	int ret = 0;
30 
31 	if (!user_mode(regs)) {
32 		/* kprobe_running() needs smp_processor_id() */
33 		preempt_disable();
34 		if (kprobe_running() && kprobe_fault_handler(regs, fsr))
35 			ret = 1;
36 		preempt_enable();
37 	}
38 
39 	return ret;
40 }
41 #else
42 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
43 {
44 	return 0;
45 }
46 #endif
47 
48 /*
49  * This is useful to dump out the page tables associated with
50  * 'addr' in mm 'mm'.
51  */
52 void show_pte(struct mm_struct *mm, unsigned long addr)
53 {
54 	pgd_t *pgd;
55 
56 	if (!mm)
57 		mm = &init_mm;
58 
59 	printk(KERN_ALERT "pgd = %p\n", mm->pgd);
60 	pgd = pgd_offset(mm, addr);
61 	printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
62 
63 	do {
64 		pmd_t *pmd;
65 		pte_t *pte;
66 
67 		if (pgd_none(*pgd))
68 			break;
69 
70 		if (pgd_bad(*pgd)) {
71 			printk("(bad)");
72 			break;
73 		}
74 
75 		pmd = pmd_offset(pgd, addr);
76 		if (PTRS_PER_PMD != 1)
77 			printk(", *pmd=%08lx", pmd_val(*pmd));
78 
79 		if (pmd_none(*pmd))
80 			break;
81 
82 		if (pmd_bad(*pmd)) {
83 			printk("(bad)");
84 			break;
85 		}
86 
87 #ifndef CONFIG_HIGHMEM
88 		/* We must not map this if we have highmem enabled */
89 		pte = pte_offset_map(pmd, addr);
90 		printk(", *pte=%08lx", pte_val(*pte));
91 		printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
92 		pte_unmap(pte);
93 #endif
94 	} while(0);
95 
96 	printk("\n");
97 }
98 
99 /*
100  * Oops.  The kernel tried to access some page that wasn't present.
101  */
102 static void
103 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
104 		  struct pt_regs *regs)
105 {
106 	/*
107 	 * Are we prepared to handle this kernel fault?
108 	 */
109 	if (fixup_exception(regs))
110 		return;
111 
112 	/*
113 	 * No handler, we'll have to terminate things with extreme prejudice.
114 	 */
115 	bust_spinlocks(1);
116 	printk(KERN_ALERT
117 		"Unable to handle kernel %s at virtual address %08lx\n",
118 		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
119 		"paging request", addr);
120 
121 	show_pte(mm, addr);
122 	die("Oops", regs, fsr);
123 	bust_spinlocks(0);
124 	do_exit(SIGKILL);
125 }
126 
127 /*
128  * Something tried to access memory that isn't in our memory map..
129  * User mode accesses just cause a SIGSEGV
130  */
131 static void
132 __do_user_fault(struct task_struct *tsk, unsigned long addr,
133 		unsigned int fsr, unsigned int sig, int code,
134 		struct pt_regs *regs)
135 {
136 	struct siginfo si;
137 
138 #ifdef CONFIG_DEBUG_USER
139 	if (user_debug & UDBG_SEGV) {
140 		printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
141 		       tsk->comm, sig, addr, fsr);
142 		show_pte(tsk->mm, addr);
143 		show_regs(regs);
144 	}
145 #endif
146 
147 	tsk->thread.address = addr;
148 	tsk->thread.error_code = fsr;
149 	tsk->thread.trap_no = 14;
150 	si.si_signo = sig;
151 	si.si_errno = 0;
152 	si.si_code = code;
153 	si.si_addr = (void __user *)addr;
154 	force_sig_info(sig, &si, tsk);
155 }
156 
157 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
158 {
159 	struct task_struct *tsk = current;
160 	struct mm_struct *mm = tsk->active_mm;
161 
162 	/*
163 	 * If we are in kernel mode at this point, we
164 	 * have no context to handle this fault with.
165 	 */
166 	if (user_mode(regs))
167 		__do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
168 	else
169 		__do_kernel_fault(mm, addr, fsr, regs);
170 }
171 
172 #define VM_FAULT_BADMAP		0x010000
173 #define VM_FAULT_BADACCESS	0x020000
174 
175 static int
176 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
177 		struct task_struct *tsk)
178 {
179 	struct vm_area_struct *vma;
180 	int fault, mask;
181 
182 	vma = find_vma(mm, addr);
183 	fault = VM_FAULT_BADMAP;
184 	if (!vma)
185 		goto out;
186 	if (vma->vm_start > addr)
187 		goto check_stack;
188 
189 	/*
190 	 * Ok, we have a good vm_area for this
191 	 * memory access, so we can handle it.
192 	 */
193 good_area:
194 	if (fsr & (1 << 11)) /* write? */
195 		mask = VM_WRITE;
196 	else
197 		mask = VM_READ|VM_EXEC|VM_WRITE;
198 
199 	fault = VM_FAULT_BADACCESS;
200 	if (!(vma->vm_flags & mask))
201 		goto out;
202 
203 	/*
204 	 * If for any reason at all we couldn't handle
205 	 * the fault, make sure we exit gracefully rather
206 	 * than endlessly redo the fault.
207 	 */
208 survive:
209 	fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, fsr & (1 << 11));
210 	if (unlikely(fault & VM_FAULT_ERROR)) {
211 		if (fault & VM_FAULT_OOM)
212 			goto out_of_memory;
213 		else if (fault & VM_FAULT_SIGBUS)
214 			return fault;
215 		BUG();
216 	}
217 	if (fault & VM_FAULT_MAJOR)
218 		tsk->maj_flt++;
219 	else
220 		tsk->min_flt++;
221 	return fault;
222 
223 out_of_memory:
224 	if (!is_global_init(tsk))
225 		goto out;
226 
227 	/*
228 	 * If we are out of memory for pid1, sleep for a while and retry
229 	 */
230 	up_read(&mm->mmap_sem);
231 	yield();
232 	down_read(&mm->mmap_sem);
233 	goto survive;
234 
235 check_stack:
236 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
237 		goto good_area;
238 out:
239 	return fault;
240 }
241 
242 static int __kprobes
243 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
244 {
245 	struct task_struct *tsk;
246 	struct mm_struct *mm;
247 	int fault, sig, code;
248 
249 	if (notify_page_fault(regs, fsr))
250 		return 0;
251 
252 	tsk = current;
253 	mm  = tsk->mm;
254 
255 	/*
256 	 * If we're in an interrupt or have no user
257 	 * context, we must not take the fault..
258 	 */
259 	if (in_atomic() || !mm)
260 		goto no_context;
261 
262 	/*
263 	 * As per x86, we may deadlock here.  However, since the kernel only
264 	 * validly references user space from well defined areas of the code,
265 	 * we can bug out early if this is from code which shouldn't.
266 	 */
267 	if (!down_read_trylock(&mm->mmap_sem)) {
268 		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
269 			goto no_context;
270 		down_read(&mm->mmap_sem);
271 	}
272 
273 	fault = __do_page_fault(mm, addr, fsr, tsk);
274 	up_read(&mm->mmap_sem);
275 
276 	/*
277 	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
278 	 */
279 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
280 		return 0;
281 
282 	/*
283 	 * If we are in kernel mode at this point, we
284 	 * have no context to handle this fault with.
285 	 */
286 	if (!user_mode(regs))
287 		goto no_context;
288 
289 	if (fault & VM_FAULT_OOM) {
290 		/*
291 		 * We ran out of memory, or some other thing
292 		 * happened to us that made us unable to handle
293 		 * the page fault gracefully.
294 		 */
295 		printk("VM: killing process %s\n", tsk->comm);
296 		do_group_exit(SIGKILL);
297 		return 0;
298 	}
299 	if (fault & VM_FAULT_SIGBUS) {
300 		/*
301 		 * We had some memory, but were unable to
302 		 * successfully fix up this page fault.
303 		 */
304 		sig = SIGBUS;
305 		code = BUS_ADRERR;
306 	} else {
307 		/*
308 		 * Something tried to access memory that
309 		 * isn't in our memory map..
310 		 */
311 		sig = SIGSEGV;
312 		code = fault == VM_FAULT_BADACCESS ?
313 			SEGV_ACCERR : SEGV_MAPERR;
314 	}
315 
316 	__do_user_fault(tsk, addr, fsr, sig, code, regs);
317 	return 0;
318 
319 no_context:
320 	__do_kernel_fault(mm, addr, fsr, regs);
321 	return 0;
322 }
323 
324 /*
325  * First Level Translation Fault Handler
326  *
327  * We enter here because the first level page table doesn't contain
328  * a valid entry for the address.
329  *
330  * If the address is in kernel space (>= TASK_SIZE), then we are
331  * probably faulting in the vmalloc() area.
332  *
333  * If the init_task's first level page tables contains the relevant
334  * entry, we copy the it to this task.  If not, we send the process
335  * a signal, fixup the exception, or oops the kernel.
336  *
337  * NOTE! We MUST NOT take any locks for this case. We may be in an
338  * interrupt or a critical region, and should only copy the information
339  * from the master page table, nothing more.
340  */
341 static int __kprobes
342 do_translation_fault(unsigned long addr, unsigned int fsr,
343 		     struct pt_regs *regs)
344 {
345 	unsigned int index;
346 	pgd_t *pgd, *pgd_k;
347 	pmd_t *pmd, *pmd_k;
348 
349 	if (addr < TASK_SIZE)
350 		return do_page_fault(addr, fsr, regs);
351 
352 	index = pgd_index(addr);
353 
354 	/*
355 	 * FIXME: CP15 C1 is write only on ARMv3 architectures.
356 	 */
357 	pgd = cpu_get_pgd() + index;
358 	pgd_k = init_mm.pgd + index;
359 
360 	if (pgd_none(*pgd_k))
361 		goto bad_area;
362 
363 	if (!pgd_present(*pgd))
364 		set_pgd(pgd, *pgd_k);
365 
366 	pmd_k = pmd_offset(pgd_k, addr);
367 	pmd   = pmd_offset(pgd, addr);
368 
369 	if (pmd_none(*pmd_k))
370 		goto bad_area;
371 
372 	copy_pmd(pmd, pmd_k);
373 	return 0;
374 
375 bad_area:
376 	do_bad_area(addr, fsr, regs);
377 	return 0;
378 }
379 
380 /*
381  * Some section permission faults need to be handled gracefully.
382  * They can happen due to a __{get,put}_user during an oops.
383  */
384 static int
385 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
386 {
387 	do_bad_area(addr, fsr, regs);
388 	return 0;
389 }
390 
391 /*
392  * This abort handler always returns "fault".
393  */
394 static int
395 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
396 {
397 	return 1;
398 }
399 
400 static struct fsr_info {
401 	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
402 	int	sig;
403 	int	code;
404 	const char *name;
405 } fsr_info[] = {
406 	/*
407 	 * The following are the standard ARMv3 and ARMv4 aborts.  ARMv5
408 	 * defines these to be "precise" aborts.
409 	 */
410 	{ do_bad,		SIGSEGV, 0,		"vector exception"		   },
411 	{ do_bad,		SIGILL,	 BUS_ADRALN,	"alignment exception"		   },
412 	{ do_bad,		SIGKILL, 0,		"terminal exception"		   },
413 	{ do_bad,		SIGILL,	 BUS_ADRALN,	"alignment exception"		   },
414 	{ do_bad,		SIGBUS,	 0,		"external abort on linefetch"	   },
415 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"section translation fault"	   },
416 	{ do_bad,		SIGBUS,	 0,		"external abort on linefetch"	   },
417 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"page translation fault"	   },
418 	{ do_bad,		SIGBUS,	 0,		"external abort on non-linefetch"  },
419 	{ do_bad,		SIGSEGV, SEGV_ACCERR,	"section domain fault"		   },
420 	{ do_bad,		SIGBUS,	 0,		"external abort on non-linefetch"  },
421 	{ do_bad,		SIGSEGV, SEGV_ACCERR,	"page domain fault"		   },
422 	{ do_bad,		SIGBUS,	 0,		"external abort on translation"	   },
423 	{ do_sect_fault,	SIGSEGV, SEGV_ACCERR,	"section permission fault"	   },
424 	{ do_bad,		SIGBUS,	 0,		"external abort on translation"	   },
425 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"page permission fault"		   },
426 	/*
427 	 * The following are "imprecise" aborts, which are signalled by bit
428 	 * 10 of the FSR, and may not be recoverable.  These are only
429 	 * supported if the CPU abort handler supports bit 10.
430 	 */
431 	{ do_bad,		SIGBUS,  0,		"unknown 16"			   },
432 	{ do_bad,		SIGBUS,  0,		"unknown 17"			   },
433 	{ do_bad,		SIGBUS,  0,		"unknown 18"			   },
434 	{ do_bad,		SIGBUS,  0,		"unknown 19"			   },
435 	{ do_bad,		SIGBUS,  0,		"lock abort"			   }, /* xscale */
436 	{ do_bad,		SIGBUS,  0,		"unknown 21"			   },
437 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"imprecise external abort"	   }, /* xscale */
438 	{ do_bad,		SIGBUS,  0,		"unknown 23"			   },
439 	{ do_bad,		SIGBUS,  0,		"dcache parity error"		   }, /* xscale */
440 	{ do_bad,		SIGBUS,  0,		"unknown 25"			   },
441 	{ do_bad,		SIGBUS,  0,		"unknown 26"			   },
442 	{ do_bad,		SIGBUS,  0,		"unknown 27"			   },
443 	{ do_bad,		SIGBUS,  0,		"unknown 28"			   },
444 	{ do_bad,		SIGBUS,  0,		"unknown 29"			   },
445 	{ do_bad,		SIGBUS,  0,		"unknown 30"			   },
446 	{ do_bad,		SIGBUS,  0,		"unknown 31"			   }
447 };
448 
449 void __init
450 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
451 		int sig, const char *name)
452 {
453 	if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
454 		fsr_info[nr].fn   = fn;
455 		fsr_info[nr].sig  = sig;
456 		fsr_info[nr].name = name;
457 	}
458 }
459 
460 /*
461  * Dispatch a data abort to the relevant handler.
462  */
463 asmlinkage void __exception
464 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
465 {
466 	const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
467 	struct siginfo info;
468 
469 	if (!inf->fn(addr, fsr, regs))
470 		return;
471 
472 	printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
473 		inf->name, fsr, addr);
474 
475 	info.si_signo = inf->sig;
476 	info.si_errno = 0;
477 	info.si_code  = inf->code;
478 	info.si_addr  = (void __user *)addr;
479 	arm_notify_die("", regs, &info, fsr, 0);
480 }
481 
482 asmlinkage void __exception
483 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
484 {
485 	do_translation_fault(addr, 0, regs);
486 }
487 
488