ARM: Update page fault handling for new OOM techniques
[deliverable/linux.git] / arch / arm / mm / fault.c
CommitLineData
1da177e4
LT
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 */
1da177e4
LT
11#include <linux/module.h>
12#include <linux/signal.h>
1da177e4 13#include <linux/mm.h>
67306da6 14#include <linux/hardirq.h>
1da177e4 15#include <linux/init.h>
25ce1dd7 16#include <linux/kprobes.h>
33fa9b13 17#include <linux/uaccess.h>
252d4c27 18#include <linux/page-flags.h>
412bb0a6 19#include <linux/sched.h>
65cec8e3 20#include <linux/highmem.h>
1da177e4
LT
21
22#include <asm/system.h>
23#include <asm/pgtable.h>
24#include <asm/tlbflush.h>
1da177e4
LT
25
26#include "fault.h"
27
c88d6aa7
RK
28/*
29 * Fault status register encodings
30 */
31#define FSR_WRITE (1 << 11)
32#define FSR_FS4 (1 << 10)
33#define FSR_FS3_0 (15)
34
35static inline int fsr_fs(unsigned int fsr)
36{
37 return (fsr & FSR_FS3_0) | (fsr & FSR_FS4) >> 6;
38}
39
09529f7a 40#ifdef CONFIG_MMU
25ce1dd7
NP
41
42#ifdef CONFIG_KPROBES
43static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
44{
45 int ret = 0;
46
47 if (!user_mode(regs)) {
48 /* kprobe_running() needs smp_processor_id() */
49 preempt_disable();
50 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
51 ret = 1;
52 preempt_enable();
53 }
54
55 return ret;
56}
57#else
58static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
59{
60 return 0;
61}
62#endif
63
1da177e4
LT
64/*
65 * This is useful to dump out the page tables associated with
66 * 'addr' in mm 'mm'.
67 */
68void show_pte(struct mm_struct *mm, unsigned long addr)
69{
70 pgd_t *pgd;
71
72 if (!mm)
73 mm = &init_mm;
74
75 printk(KERN_ALERT "pgd = %p\n", mm->pgd);
76 pgd = pgd_offset(mm, addr);
77 printk(KERN_ALERT "[%08lx] *pgd=%08lx", addr, pgd_val(*pgd));
78
79 do {
80 pmd_t *pmd;
81 pte_t *pte;
82
83 if (pgd_none(*pgd))
84 break;
85
86 if (pgd_bad(*pgd)) {
87 printk("(bad)");
88 break;
89 }
90
91 pmd = pmd_offset(pgd, addr);
da46c79a
NP
92 if (PTRS_PER_PMD != 1)
93 printk(", *pmd=%08lx", pmd_val(*pmd));
1da177e4
LT
94
95 if (pmd_none(*pmd))
96 break;
97
98 if (pmd_bad(*pmd)) {
99 printk("(bad)");
100 break;
101 }
102
1da177e4 103 /* We must not map this if we have highmem enabled */
252d4c27
NP
104 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
105 break;
106
1da177e4
LT
107 pte = pte_offset_map(pmd, addr);
108 printk(", *pte=%08lx", pte_val(*pte));
109 printk(", *ppte=%08lx", pte_val(pte[-PTRS_PER_PTE]));
110 pte_unmap(pte);
1da177e4
LT
111 } while(0);
112
113 printk("\n");
114}
09529f7a
CM
115#else /* CONFIG_MMU */
116void show_pte(struct mm_struct *mm, unsigned long addr)
117{ }
118#endif /* CONFIG_MMU */
1da177e4
LT
119
120/*
121 * Oops. The kernel tried to access some page that wasn't present.
122 */
123static void
124__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
125 struct pt_regs *regs)
126{
127 /*
128 * Are we prepared to handle this kernel fault?
129 */
130 if (fixup_exception(regs))
131 return;
132
133 /*
134 * No handler, we'll have to terminate things with extreme prejudice.
135 */
136 bust_spinlocks(1);
137 printk(KERN_ALERT
138 "Unable to handle kernel %s at virtual address %08lx\n",
139 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
140 "paging request", addr);
141
142 show_pte(mm, addr);
143 die("Oops", regs, fsr);
144 bust_spinlocks(0);
145 do_exit(SIGKILL);
146}
147
148/*
149 * Something tried to access memory that isn't in our memory map..
150 * User mode accesses just cause a SIGSEGV
151 */
152static void
153__do_user_fault(struct task_struct *tsk, unsigned long addr,
2d137c24 154 unsigned int fsr, unsigned int sig, int code,
155 struct pt_regs *regs)
1da177e4
LT
156{
157 struct siginfo si;
158
159#ifdef CONFIG_DEBUG_USER
160 if (user_debug & UDBG_SEGV) {
2d137c24 161 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
162 tsk->comm, sig, addr, fsr);
1da177e4
LT
163 show_pte(tsk->mm, addr);
164 show_regs(regs);
165 }
166#endif
167
168 tsk->thread.address = addr;
169 tsk->thread.error_code = fsr;
170 tsk->thread.trap_no = 14;
2d137c24 171 si.si_signo = sig;
1da177e4
LT
172 si.si_errno = 0;
173 si.si_code = code;
174 si.si_addr = (void __user *)addr;
2d137c24 175 force_sig_info(sig, &si, tsk);
1da177e4
LT
176}
177
e5beac37 178void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
1da177e4 179{
e5beac37
RK
180 struct task_struct *tsk = current;
181 struct mm_struct *mm = tsk->active_mm;
182
1da177e4
LT
183 /*
184 * If we are in kernel mode at this point, we
185 * have no context to handle this fault with.
186 */
187 if (user_mode(regs))
2d137c24 188 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
1da177e4
LT
189 else
190 __do_kernel_fault(mm, addr, fsr, regs);
191}
192
09529f7a 193#ifdef CONFIG_MMU
5c72fc5c
NP
194#define VM_FAULT_BADMAP 0x010000
195#define VM_FAULT_BADACCESS 0x020000
1da177e4
LT
196
197static int
198__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
199 struct task_struct *tsk)
200{
201 struct vm_area_struct *vma;
202 int fault, mask;
203
204 vma = find_vma(mm, addr);
205 fault = VM_FAULT_BADMAP;
206 if (!vma)
207 goto out;
208 if (vma->vm_start > addr)
209 goto check_stack;
210
211 /*
212 * Ok, we have a good vm_area for this
213 * memory access, so we can handle it.
214 */
215good_area:
c88d6aa7 216 if (fsr & FSR_WRITE)
1da177e4
LT
217 mask = VM_WRITE;
218 else
df67b3da 219 mask = VM_READ|VM_EXEC|VM_WRITE;
1da177e4
LT
220
221 fault = VM_FAULT_BADACCESS;
222 if (!(vma->vm_flags & mask))
223 goto out;
224
225 /*
b42c6344
RK
226 * If for any reason at all we couldn't handle the fault, make
227 * sure we exit gracefully rather than endlessly redo the fault.
1da177e4 228 */
c88d6aa7 229 fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & FSR_WRITE) ? FAULT_FLAG_WRITE : 0);
b42c6344
RK
230 if (unlikely(fault & VM_FAULT_ERROR))
231 return fault;
83c54070 232 if (fault & VM_FAULT_MAJOR)
1da177e4 233 tsk->maj_flt++;
83c54070 234 else
1da177e4 235 tsk->min_flt++;
83c54070 236 return fault;
1da177e4 237
1da177e4
LT
238check_stack:
239 if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
240 goto good_area;
241out:
242 return fault;
243}
244
785d3cd2 245static int __kprobes
1da177e4
LT
246do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
247{
248 struct task_struct *tsk;
249 struct mm_struct *mm;
2d137c24 250 int fault, sig, code;
1da177e4 251
25ce1dd7
NP
252 if (notify_page_fault(regs, fsr))
253 return 0;
254
1da177e4
LT
255 tsk = current;
256 mm = tsk->mm;
257
258 /*
259 * If we're in an interrupt or have no user
260 * context, we must not take the fault..
261 */
6edaf68a 262 if (in_atomic() || !mm)
1da177e4
LT
263 goto no_context;
264
840ff6a4
RK
265 /*
266 * As per x86, we may deadlock here. However, since the kernel only
267 * validly references user space from well defined areas of the code,
268 * we can bug out early if this is from code which shouldn't.
269 */
270 if (!down_read_trylock(&mm->mmap_sem)) {
271 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
272 goto no_context;
273 down_read(&mm->mmap_sem);
274 }
275
1da177e4
LT
276 fault = __do_page_fault(mm, addr, fsr, tsk);
277 up_read(&mm->mmap_sem);
278
279 /*
ff2afb9d 280 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
1da177e4 281 */
5c72fc5c 282 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
1da177e4
LT
283 return 0;
284
b42c6344
RK
285 if (fault & VM_FAULT_OOM) {
286 /*
287 * We ran out of memory, call the OOM killer, and return to
288 * userspace (which will retry the fault, or kill us if we
289 * got oom-killed)
290 */
291 pagefault_out_of_memory();
292 return 0;
293 }
294
1da177e4
LT
295 /*
296 * If we are in kernel mode at this point, we
297 * have no context to handle this fault with.
298 */
299 if (!user_mode(regs))
300 goto no_context;
301
83c54070 302 if (fault & VM_FAULT_SIGBUS) {
2d137c24 303 /*
304 * We had some memory, but were unable to
305 * successfully fix up this page fault.
306 */
307 sig = SIGBUS;
308 code = BUS_ADRERR;
83c54070 309 } else {
2d137c24 310 /*
311 * Something tried to access memory that
312 * isn't in our memory map..
313 */
314 sig = SIGSEGV;
315 code = fault == VM_FAULT_BADACCESS ?
316 SEGV_ACCERR : SEGV_MAPERR;
1da177e4 317 }
1da177e4 318
2d137c24 319 __do_user_fault(tsk, addr, fsr, sig, code, regs);
320 return 0;
1da177e4
LT
321
322no_context:
323 __do_kernel_fault(mm, addr, fsr, regs);
324 return 0;
325}
09529f7a
CM
326#else /* CONFIG_MMU */
327static int
328do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
329{
330 return 0;
331}
332#endif /* CONFIG_MMU */
1da177e4
LT
333
334/*
335 * First Level Translation Fault Handler
336 *
337 * We enter here because the first level page table doesn't contain
338 * a valid entry for the address.
339 *
340 * If the address is in kernel space (>= TASK_SIZE), then we are
341 * probably faulting in the vmalloc() area.
342 *
343 * If the init_task's first level page tables contains the relevant
344 * entry, we copy the it to this task. If not, we send the process
345 * a signal, fixup the exception, or oops the kernel.
346 *
347 * NOTE! We MUST NOT take any locks for this case. We may be in an
348 * interrupt or a critical region, and should only copy the information
349 * from the master page table, nothing more.
350 */
09529f7a 351#ifdef CONFIG_MMU
785d3cd2 352static int __kprobes
1da177e4
LT
353do_translation_fault(unsigned long addr, unsigned int fsr,
354 struct pt_regs *regs)
355{
1da177e4
LT
356 unsigned int index;
357 pgd_t *pgd, *pgd_k;
358 pmd_t *pmd, *pmd_k;
359
360 if (addr < TASK_SIZE)
361 return do_page_fault(addr, fsr, regs);
362
363 index = pgd_index(addr);
364
365 /*
366 * FIXME: CP15 C1 is write only on ARMv3 architectures.
367 */
368 pgd = cpu_get_pgd() + index;
369 pgd_k = init_mm.pgd + index;
370
371 if (pgd_none(*pgd_k))
372 goto bad_area;
373
374 if (!pgd_present(*pgd))
375 set_pgd(pgd, *pgd_k);
376
377 pmd_k = pmd_offset(pgd_k, addr);
378 pmd = pmd_offset(pgd, addr);
379
380 if (pmd_none(*pmd_k))
381 goto bad_area;
382
383 copy_pmd(pmd, pmd_k);
384 return 0;
385
386bad_area:
e5beac37 387 do_bad_area(addr, fsr, regs);
1da177e4
LT
388 return 0;
389}
09529f7a
CM
390#else /* CONFIG_MMU */
391static int
392do_translation_fault(unsigned long addr, unsigned int fsr,
393 struct pt_regs *regs)
394{
395 return 0;
396}
397#endif /* CONFIG_MMU */
1da177e4
LT
398
399/*
400 * Some section permission faults need to be handled gracefully.
401 * They can happen due to a __{get,put}_user during an oops.
402 */
403static int
404do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
405{
e5beac37 406 do_bad_area(addr, fsr, regs);
1da177e4
LT
407 return 0;
408}
409
410/*
411 * This abort handler always returns "fault".
412 */
413static int
414do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
415{
416 return 1;
417}
418
419static struct fsr_info {
420 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
421 int sig;
cfb0810e 422 int code;
1da177e4
LT
423 const char *name;
424} fsr_info[] = {
425 /*
426 * The following are the standard ARMv3 and ARMv4 aborts. ARMv5
427 * defines these to be "precise" aborts.
428 */
cfb0810e
RK
429 { do_bad, SIGSEGV, 0, "vector exception" },
430 { do_bad, SIGILL, BUS_ADRALN, "alignment exception" },
431 { do_bad, SIGKILL, 0, "terminal exception" },
432 { do_bad, SIGILL, BUS_ADRALN, "alignment exception" },
433 { do_bad, SIGBUS, 0, "external abort on linefetch" },
434 { do_translation_fault, SIGSEGV, SEGV_MAPERR, "section translation fault" },
435 { do_bad, SIGBUS, 0, "external abort on linefetch" },
436 { do_page_fault, SIGSEGV, SEGV_MAPERR, "page translation fault" },
437 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
438 { do_bad, SIGSEGV, SEGV_ACCERR, "section domain fault" },
439 { do_bad, SIGBUS, 0, "external abort on non-linefetch" },
440 { do_bad, SIGSEGV, SEGV_ACCERR, "page domain fault" },
441 { do_bad, SIGBUS, 0, "external abort on translation" },
442 { do_sect_fault, SIGSEGV, SEGV_ACCERR, "section permission fault" },
443 { do_bad, SIGBUS, 0, "external abort on translation" },
444 { do_page_fault, SIGSEGV, SEGV_ACCERR, "page permission fault" },
1da177e4
LT
445 /*
446 * The following are "imprecise" aborts, which are signalled by bit
447 * 10 of the FSR, and may not be recoverable. These are only
448 * supported if the CPU abort handler supports bit 10.
449 */
cfb0810e
RK
450 { do_bad, SIGBUS, 0, "unknown 16" },
451 { do_bad, SIGBUS, 0, "unknown 17" },
452 { do_bad, SIGBUS, 0, "unknown 18" },
453 { do_bad, SIGBUS, 0, "unknown 19" },
454 { do_bad, SIGBUS, 0, "lock abort" }, /* xscale */
455 { do_bad, SIGBUS, 0, "unknown 21" },
456 { do_bad, SIGBUS, BUS_OBJERR, "imprecise external abort" }, /* xscale */
457 { do_bad, SIGBUS, 0, "unknown 23" },
458 { do_bad, SIGBUS, 0, "dcache parity error" }, /* xscale */
459 { do_bad, SIGBUS, 0, "unknown 25" },
460 { do_bad, SIGBUS, 0, "unknown 26" },
461 { do_bad, SIGBUS, 0, "unknown 27" },
462 { do_bad, SIGBUS, 0, "unknown 28" },
463 { do_bad, SIGBUS, 0, "unknown 29" },
464 { do_bad, SIGBUS, 0, "unknown 30" },
465 { do_bad, SIGBUS, 0, "unknown 31" }
1da177e4
LT
466};
467
468void __init
469hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
470 int sig, const char *name)
471{
472 if (nr >= 0 && nr < ARRAY_SIZE(fsr_info)) {
473 fsr_info[nr].fn = fn;
474 fsr_info[nr].sig = sig;
475 fsr_info[nr].name = name;
476 }
477}
478
479/*
480 * Dispatch a data abort to the relevant handler.
481 */
7ab3f8d5 482asmlinkage void __exception
1da177e4
LT
483do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
484{
c88d6aa7 485 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
cfb0810e 486 struct siginfo info;
1da177e4
LT
487
488 if (!inf->fn(addr, fsr, regs))
489 return;
490
491 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
492 inf->name, fsr, addr);
cfb0810e
RK
493
494 info.si_signo = inf->sig;
495 info.si_errno = 0;
496 info.si_code = inf->code;
497 info.si_addr = (void __user *)addr;
1eeb66a1 498 arm_notify_die("", regs, &info, fsr, 0);
1da177e4
LT
499}
500
7ab3f8d5 501asmlinkage void __exception
1da177e4
LT
502do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
503{
504 do_translation_fault(addr, 0, regs);
505}
506
This page took 0.460067 seconds and 5 git commands to generate.