xref: /openbmc/linux/arch/arm/mm/fault.c (revision 367b8112)
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/init.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 
18 #include <asm/system.h>
19 #include <asm/pgtable.h>
20 #include <asm/tlbflush.h>
21 
22 #include "fault.h"
23 
24 
25 #ifdef CONFIG_KPROBES
26 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
27 {
28 	int ret = 0;
29 
30 	if (!user_mode(regs)) {
31 		/* kprobe_running() needs smp_processor_id() */
32 		preempt_disable();
33 		if (kprobe_running() && kprobe_fault_handler(regs, fsr))
34 			ret = 1;
35 		preempt_enable();
36 	}
37 
38 	return ret;
39 }
40 #else
41 static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
42 {
43 	return 0;
44 }
45 #endif
46 
47 /*
48  * This is useful to dump out the page tables associated with
49  * 'addr' in mm 'mm'.
50  */
51 void show_pte(struct mm_struct *mm, unsigned long addr)
52 {
53 	pgd_t *pgd;
54 
55 	if (!mm)
56 		mm = &init_mm;
57 
58 	printk(KERN_ALERT "pgd = %p\n", mm->pgd);
59 	pgd = pgd_offset(mm, addr);
60 	printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
61 
62 	do {
63 		pmd_t *pmd;
64 		pte_t *pte;
65 
66 		if (pgd_none(*pgd))
67 			break;
68 
69 		if (pgd_bad(*pgd)) {
70 			printk("(bad)");
71 			break;
72 		}
73 
74 		pmd = pmd_offset(pgd, addr);
75 		if (PTRS_PER_PMD != 1)
76 			printk(", *pmd=%08lx", pmd_val(*pmd));
77 
78 		if (pmd_none(*pmd))
79 			break;
80 
81 		if (pmd_bad(*pmd)) {
82 			printk("(bad)");
83 			break;
84 		}
85 
86 #ifndef CONFIG_HIGHMEM
87 		/* We must not map this if we have highmem enabled */
88 		pte = pte_offset_map(pmd, addr);
89 		printk(", *pte=%08lx", pte_val(*pte));
90 		printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
91 		pte_unmap(pte);
92 #endif
93 	} while(0);
94 
95 	printk("\n");
96 }
97 
98 /*
99  * Oops.  The kernel tried to access some page that wasn't present.
100  */
101 static void
102 __do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
103 		  struct pt_regs *regs)
104 {
105 	/*
106 	 * Are we prepared to handle this kernel fault?
107 	 */
108 	if (fixup_exception(regs))
109 		return;
110 
111 	/*
112 	 * No handler, we'll have to terminate things with extreme prejudice.
113 	 */
114 	bust_spinlocks(1);
115 	printk(KERN_ALERT
116 		"Unable to handle kernel %s at virtual address %08lx\n",
117 		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
118 		"paging request", addr);
119 
120 	show_pte(mm, addr);
121 	die("Oops", regs, fsr);
122 	bust_spinlocks(0);
123 	do_exit(SIGKILL);
124 }
125 
126 /*
127  * Something tried to access memory that isn't in our memory map..
128  * User mode accesses just cause a SIGSEGV
129  */
130 static void
131 __do_user_fault(struct task_struct *tsk, unsigned long addr,
132 		unsigned int fsr, unsigned int sig, int code,
133 		struct pt_regs *regs)
134 {
135 	struct siginfo si;
136 
137 #ifdef CONFIG_DEBUG_USER
138 	if (user_debug & UDBG_SEGV) {
139 		printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
140 		       tsk->comm, sig, addr, fsr);
141 		show_pte(tsk->mm, addr);
142 		show_regs(regs);
143 	}
144 #endif
145 
146 	tsk->thread.address = addr;
147 	tsk->thread.error_code = fsr;
148 	tsk->thread.trap_no = 14;
149 	si.si_signo = sig;
150 	si.si_errno = 0;
151 	si.si_code = code;
152 	si.si_addr = (void __user *)addr;
153 	force_sig_info(sig, &si, tsk);
154 }
155 
156 void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
157 {
158 	struct task_struct *tsk = current;
159 	struct mm_struct *mm = tsk->active_mm;
160 
161 	/*
162 	 * If we are in kernel mode at this point, we
163 	 * have no context to handle this fault with.
164 	 */
165 	if (user_mode(regs))
166 		__do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
167 	else
168 		__do_kernel_fault(mm, addr, fsr, regs);
169 }
170 
171 #define VM_FAULT_BADMAP		0x010000
172 #define VM_FAULT_BADACCESS	0x020000
173 
174 static int
175 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
176 		struct task_struct *tsk)
177 {
178 	struct vm_area_struct *vma;
179 	int fault, mask;
180 
181 	vma = find_vma(mm, addr);
182 	fault = VM_FAULT_BADMAP;
183 	if (!vma)
184 		goto out;
185 	if (vma->vm_start > addr)
186 		goto check_stack;
187 
188 	/*
189 	 * Ok, we have a good vm_area for this
190 	 * memory access, so we can handle it.
191 	 */
192 good_area:
193 	if (fsr & (1 << 11)) /* write? */
194 		mask = VM_WRITE;
195 	else
196 		mask = VM_READ|VM_EXEC|VM_WRITE;
197 
198 	fault = VM_FAULT_BADACCESS;
199 	if (!(vma->vm_flags & mask))
200 		goto out;
201 
202 	/*
203 	 * If for any reason at all we couldn't handle
204 	 * the fault, make sure we exit gracefully rather
205 	 * than endlessly redo the fault.
206 	 */
207 survive:
208 	fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, fsr & (1 << 11));
209 	if (unlikely(fault & VM_FAULT_ERROR)) {
210 		if (fault & VM_FAULT_OOM)
211 			goto out_of_memory;
212 		else if (fault & VM_FAULT_SIGBUS)
213 			return fault;
214 		BUG();
215 	}
216 	if (fault & VM_FAULT_MAJOR)
217 		tsk->maj_flt++;
218 	else
219 		tsk->min_flt++;
220 	return fault;
221 
222 out_of_memory:
223 	if (!is_global_init(tsk))
224 		goto out;
225 
226 	/*
227 	 * If we are out of memory for pid1, sleep for a while and retry
228 	 */
229 	up_read(&mm->mmap_sem);
230 	yield();
231 	down_read(&mm->mmap_sem);
232 	goto survive;
233 
234 check_stack:
235 	if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
236 		goto good_area;
237 out:
238 	return fault;
239 }
240 
241 static int __kprobes
242 do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
243 {
244 	struct task_struct *tsk;
245 	struct mm_struct *mm;
246 	int fault, sig, code;
247 
248 	if (notify_page_fault(regs, fsr))
249 		return 0;
250 
251 	tsk = current;
252 	mm  = tsk->mm;
253 
254 	/*
255 	 * If we're in an interrupt or have no user
256 	 * context, we must not take the fault..
257 	 */
258 	if (in_atomic() || !mm)
259 		goto no_context;
260 
261 	/*
262 	 * As per x86, we may deadlock here.  However, since the kernel only
263 	 * validly references user space from well defined areas of the code,
264 	 * we can bug out early if this is from code which shouldn't.
265 	 */
266 	if (!down_read_trylock(&mm->mmap_sem)) {
267 		if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
268 			goto no_context;
269 		down_read(&mm->mmap_sem);
270 	}
271 
272 	fault = __do_page_fault(mm, addr, fsr, tsk);
273 	up_read(&mm->mmap_sem);
274 
275 	/*
276 	 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
277 	 */
278 	if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
279 		return 0;
280 
281 	/*
282 	 * If we are in kernel mode at this point, we
283 	 * have no context to handle this fault with.
284 	 */
285 	if (!user_mode(regs))
286 		goto no_context;
287 
288 	if (fault & VM_FAULT_OOM) {
289 		/*
290 		 * We ran out of memory, or some other thing
291 		 * happened to us that made us unable to handle
292 		 * the page fault gracefully.
293 		 */
294 		printk("VM: killing process %s\n", tsk->comm);
295 		do_group_exit(SIGKILL);
296 		return 0;
297 	}
298 	if (fault & VM_FAULT_SIGBUS) {
299 		/*
300 		 * We had some memory, but were unable to
301 		 * successfully fix up this page fault.
302 		 */
303 		sig = SIGBUS;
304 		code = BUS_ADRERR;
305 	} else {
306 		/*
307 		 * Something tried to access memory that
308 		 * isn't in our memory map..
309 		 */
310 		sig = SIGSEGV;
311 		code = fault == VM_FAULT_BADACCESS ?
312 			SEGV_ACCERR : SEGV_MAPERR;
313 	}
314 
315 	__do_user_fault(tsk, addr, fsr, sig, code, regs);
316 	return 0;
317 
318 no_context:
319 	__do_kernel_fault(mm, addr, fsr, regs);
320 	return 0;
321 }
322 
323 /*
324  * First Level Translation Fault Handler
325  *
326  * We enter here because the first level page table doesn't contain
327  * a valid entry for the address.
328  *
329  * If the address is in kernel space (>= TASK_SIZE), then we are
330  * probably faulting in the vmalloc() area.
331  *
332  * If the init_task's first level page tables contains the relevant
333  * entry, we copy the it to this task.  If not, we send the process
334  * a signal, fixup the exception, or oops the kernel.
335  *
336  * NOTE! We MUST NOT take any locks for this case. We may be in an
337  * interrupt or a critical region, and should only copy the information
338  * from the master page table, nothing more.
339  */
340 static int __kprobes
341 do_translation_fault(unsigned long addr, unsigned int fsr,
342 		     struct pt_regs *regs)
343 {
344 	unsigned int index;
345 	pgd_t *pgd, *pgd_k;
346 	pmd_t *pmd, *pmd_k;
347 
348 	if (addr < TASK_SIZE)
349 		return do_page_fault(addr, fsr, regs);
350 
351 	index = pgd_index(addr);
352 
353 	/*
354 	 * FIXME: CP15 C1 is write only on ARMv3 architectures.
355 	 */
356 	pgd = cpu_get_pgd() + index;
357 	pgd_k = init_mm.pgd + index;
358 
359 	if (pgd_none(*pgd_k))
360 		goto bad_area;
361 
362 	if (!pgd_present(*pgd))
363 		set_pgd(pgd, *pgd_k);
364 
365 	pmd_k = pmd_offset(pgd_k, addr);
366 	pmd   = pmd_offset(pgd, addr);
367 
368 	if (pmd_none(*pmd_k))
369 		goto bad_area;
370 
371 	copy_pmd(pmd, pmd_k);
372 	return 0;
373 
374 bad_area:
375 	do_bad_area(addr, fsr, regs);
376 	return 0;
377 }
378 
379 /*
380  * Some section permission faults need to be handled gracefully.
381  * They can happen due to a __{get,put}_user during an oops.
382  */
383 static int
384 do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
385 {
386 	do_bad_area(addr, fsr, regs);
387 	return 0;
388 }
389 
390 /*
391  * This abort handler always returns "fault".
392  */
393 static int
394 do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
395 {
396 	return 1;
397 }
398 
399 static struct fsr_info {
400 	int	(*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
401 	int	sig;
402 	int	code;
403 	const char *name;
404 } fsr_info[] = {
405 	/*
406 	 * The following are the standard ARMv3 and ARMv4 aborts.  ARMv5
407 	 * defines these to be "precise" aborts.
408 	 */
409 	{ do_bad,		SIGSEGV, 0,		"vector exception"		   },
410 	{ do_bad,		SIGILL,	 BUS_ADRALN,	"alignment exception"		   },
411 	{ do_bad,		SIGKILL, 0,		"terminal exception"		   },
412 	{ do_bad,		SIGILL,	 BUS_ADRALN,	"alignment exception"		   },
413 	{ do_bad,		SIGBUS,	 0,		"external abort on linefetch"	   },
414 	{ do_translation_fault,	SIGSEGV, SEGV_MAPERR,	"section translation fault"	   },
415 	{ do_bad,		SIGBUS,	 0,		"external abort on linefetch"	   },
416 	{ do_page_fault,	SIGSEGV, SEGV_MAPERR,	"page translation fault"	   },
417 	{ do_bad,		SIGBUS,	 0,		"external abort on non-linefetch"  },
418 	{ do_bad,		SIGSEGV, SEGV_ACCERR,	"section domain fault"		   },
419 	{ do_bad,		SIGBUS,	 0,		"external abort on non-linefetch"  },
420 	{ do_bad,		SIGSEGV, SEGV_ACCERR,	"page domain fault"		   },
421 	{ do_bad,		SIGBUS,	 0,		"external abort on translation"	   },
422 	{ do_sect_fault,	SIGSEGV, SEGV_ACCERR,	"section permission fault"	   },
423 	{ do_bad,		SIGBUS,	 0,		"external abort on translation"	   },
424 	{ do_page_fault,	SIGSEGV, SEGV_ACCERR,	"page permission fault"		   },
425 	/*
426 	 * The following are "imprecise" aborts, which are signalled by bit
427 	 * 10 of the FSR, and may not be recoverable.  These are only
428 	 * supported if the CPU abort handler supports bit 10.
429 	 */
430 	{ do_bad,		SIGBUS,  0,		"unknown 16"			   },
431 	{ do_bad,		SIGBUS,  0,		"unknown 17"			   },
432 	{ do_bad,		SIGBUS,  0,		"unknown 18"			   },
433 	{ do_bad,		SIGBUS,  0,		"unknown 19"			   },
434 	{ do_bad,		SIGBUS,  0,		"lock abort"			   }, /* xscale */
435 	{ do_bad,		SIGBUS,  0,		"unknown 21"			   },
436 	{ do_bad,		SIGBUS,  BUS_OBJERR,	"imprecise external abort"	   }, /* xscale */
437 	{ do_bad,		SIGBUS,  0,		"unknown 23"			   },
438 	{ do_bad,		SIGBUS,  0,		"dcache parity error"		   }, /* xscale */
439 	{ do_bad,		SIGBUS,  0,		"unknown 25"			   },
440 	{ do_bad,		SIGBUS,  0,		"unknown 26"			   },
441 	{ do_bad,		SIGBUS,  0,		"unknown 27"			   },
442 	{ do_bad,		SIGBUS,  0,		"unknown 28"			   },
443 	{ do_bad,		SIGBUS,  0,		"unknown 29"			   },
444 	{ do_bad,		SIGBUS,  0,		"unknown 30"			   },
445 	{ do_bad,		SIGBUS,  0,		"unknown 31"			   }
446 };
447 
448 void __init
449 hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
450 		int sig, const char *name)
451 {
452 	if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
453 		fsr_info[nr].fn   = fn;
454 		fsr_info[nr].sig  = sig;
455 		fsr_info[nr].name = name;
456 	}
457 }
458 
459 /*
460  * Dispatch a data abort to the relevant handler.
461  */
462 asmlinkage void __exception
463 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
464 {
465 	const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
466 	struct siginfo info;
467 
468 	if (!inf->fn(addr, fsr, regs))
469 		return;
470 
471 	printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
472 		inf->name, fsr, addr);
473 
474 	info.si_signo = inf->sig;
475 	info.si_errno = 0;
476 	info.si_code  = inf->code;
477 	info.si_addr  = (void __user *)addr;
478 	arm_notify_die("", regs, &info, fsr, 0);
479 }
480 
481 asmlinkage void __exception
482 do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
483 {
484 	do_translation_fault(addr, 0, regs);
485 }
486 
487