atomic: implement generic atomic_dec_if_positive()
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
7ada189f 21#include <linux/perf_event.h>
1da177e4 22
5a567d78 23#include <asm/exception.h>
1da177e4 24#include <asm/pgtable.h>
9f97da78
DH
25#include <asm/system_misc.h>
26#include <asm/system_info.h>
1da177e4 27#include <asm/tlbflush.h>
1da177e4
LT
28
29#include "fault.h"
30
09529f7a 31#ifdef CONFIG_MMU
25ce1dd7
NP
32
33#ifdef CONFIG_KPROBES
34static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
35{
36 int ret = 0;
37
38 if (!user_mode(regs)) {
39 /* kprobe_running() needs smp_processor_id() */
40 preempt_disable();
41 if (kprobe_running() && kprobe_fault_handler(regs, fsr))
42 ret = 1;
43 preempt_enable();
44 }
45
46 return ret;
47}
48#else
49static inline int notify_page_fault(struct pt_regs *regs, unsigned int fsr)
50{
51 return 0;
52}
53#endif
54
1da177e4
LT
55/*
56 * This is useful to dump out the page tables associated with
57 * 'addr' in mm 'mm'.
58 */
59void show_pte(struct mm_struct *mm, unsigned long addr)
60{
61 pgd_t *pgd;
62
63 if (!mm)
64 mm = &init_mm;
65
66 printk(KERN_ALERT "pgd = %p\n", mm->pgd);
67 pgd = pgd_offset(mm, addr);
29a38193
WD
68 printk(KERN_ALERT "[%08lx] *pgd=%08llx",
69 addr, (long long)pgd_val(*pgd));
1da177e4
LT
70
71 do {
516295e5 72 pud_t *pud;
1da177e4
LT
73 pmd_t *pmd;
74 pte_t *pte;
75
76 if (pgd_none(*pgd))
77 break;
78
79 if (pgd_bad(*pgd)) {
80 printk("(bad)");
81 break;
82 }
83
516295e5
RK
84 pud = pud_offset(pgd, addr);
85 if (PTRS_PER_PUD != 1)
140d5dc1 86 printk(", *pud=%08llx", (long long)pud_val(*pud));
516295e5
RK
87
88 if (pud_none(*pud))
89 break;
90
91 if (pud_bad(*pud)) {
92 printk("(bad)");
93 break;
94 }
95
96 pmd = pmd_offset(pud, addr);
da46c79a 97 if (PTRS_PER_PMD != 1)
29a38193 98 printk(", *pmd=%08llx", (long long)pmd_val(*pmd));
1da177e4
LT
99
100 if (pmd_none(*pmd))
101 break;
102
103 if (pmd_bad(*pmd)) {
104 printk("(bad)");
105 break;
106 }
107
1da177e4 108 /* We must not map this if we have highmem enabled */
252d4c27
NP
109 if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
110 break;
111
1da177e4 112 pte = pte_offset_map(pmd, addr);
29a38193 113 printk(", *pte=%08llx", (long long)pte_val(*pte));
f7b8156d 114#ifndef CONFIG_ARM_LPAE
29a38193
WD
115 printk(", *ppte=%08llx",
116 (long long)pte_val(pte[PTE_HWTABLE_PTRS]));
f7b8156d 117#endif
1da177e4 118 pte_unmap(pte);
1da177e4
LT
119 } while(0);
120
121 printk("\n");
122}
09529f7a
CM
123#else /* CONFIG_MMU */
124void show_pte(struct mm_struct *mm, unsigned long addr)
125{ }
126#endif /* CONFIG_MMU */
1da177e4
LT
127
128/*
129 * Oops. The kernel tried to access some page that wasn't present.
130 */
131static void
132__do_kernel_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
133 struct pt_regs *regs)
134{
135 /*
136 * Are we prepared to handle this kernel fault?
137 */
138 if (fixup_exception(regs))
139 return;
140
141 /*
142 * No handler, we'll have to terminate things with extreme prejudice.
143 */
144 bust_spinlocks(1);
145 printk(KERN_ALERT
146 "Unable to handle kernel %s at virtual address %08lx\n",
147 (addr < PAGE_SIZE) ? "NULL pointer dereference" :
148 "paging request", addr);
149
150 show_pte(mm, addr);
151 die("Oops", regs, fsr);
152 bust_spinlocks(0);
153 do_exit(SIGKILL);
154}
155
156/*
157 * Something tried to access memory that isn't in our memory map..
158 * User mode accesses just cause a SIGSEGV
159 */
160static void
161__do_user_fault(struct task_struct *tsk, unsigned long addr,
2d137c24
AM
162 unsigned int fsr, unsigned int sig, int code,
163 struct pt_regs *regs)
1da177e4
LT
164{
165 struct siginfo si;
166
167#ifdef CONFIG_DEBUG_USER
f5274c2d
JM
168 if (((user_debug & UDBG_SEGV) && (sig == SIGSEGV)) ||
169 ((user_debug & UDBG_BUS) && (sig == SIGBUS))) {
2d137c24
AM
170 printk(KERN_DEBUG "%s: unhandled page fault (%d) at 0x%08lx, code 0x%03x\n",
171 tsk->comm, sig, addr, fsr);
1da177e4
LT
172 show_pte(tsk->mm, addr);
173 show_regs(regs);
174 }
175#endif
176
177 tsk->thread.address = addr;
178 tsk->thread.error_code = fsr;
179 tsk->thread.trap_no = 14;
2d137c24 180 si.si_signo = sig;
1da177e4
LT
181 si.si_errno = 0;
182 si.si_code = code;
183 si.si_addr = (void __user *)addr;
2d137c24 184 force_sig_info(sig, &si, tsk);
1da177e4
LT
185}
186
e5beac37 187void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
1da177e4 188{
e5beac37
RK
189 struct task_struct *tsk = current;
190 struct mm_struct *mm = tsk->active_mm;
191
1da177e4
LT
192 /*
193 * If we are in kernel mode at this point, we
194 * have no context to handle this fault with.
195 */
196 if (user_mode(regs))
2d137c24 197 __do_user_fault(tsk, addr, fsr, SIGSEGV, SEGV_MAPERR, regs);
1da177e4
LT
198 else
199 __do_kernel_fault(mm, addr, fsr, regs);
200}
201
09529f7a 202#ifdef CONFIG_MMU
5c72fc5c
NP
203#define VM_FAULT_BADMAP 0x010000
204#define VM_FAULT_BADACCESS 0x020000
1da177e4 205
d374bf14
RK
206/*
207 * Check that the permissions on the VMA allow for the fault which occurred.
208 * If we encountered a write fault, we must have write permission, otherwise
209 * we allow any permission.
210 */
211static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
212{
213 unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
214
215 if (fsr & FSR_WRITE)
216 mask = VM_WRITE;
df297bf6
RK
217 if (fsr & FSR_LNX_PF)
218 mask = VM_EXEC;
d374bf14
RK
219
220 return vma->vm_flags & mask ? false : true;
221}
222
223static int __kprobes
1da177e4 224__do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
8878a539 225 unsigned int flags, struct task_struct *tsk)
1da177e4
LT
226{
227 struct vm_area_struct *vma;
d374bf14 228 int fault;
1da177e4
LT
229
230 vma = find_vma(mm, addr);
231 fault = VM_FAULT_BADMAP;
d374bf14 232 if (unlikely(!vma))
1da177e4 233 goto out;
d374bf14 234 if (unlikely(vma->vm_start > addr))
1da177e4
LT
235 goto check_stack;
236
237 /*
238 * Ok, we have a good vm_area for this
239 * memory access, so we can handle it.
240 */
241good_area:
d374bf14
RK
242 if (access_error(fsr, vma)) {
243 fault = VM_FAULT_BADACCESS;
1da177e4 244 goto out;
d374bf14 245 }
1da177e4 246
8878a539 247 return handle_mm_fault(mm, vma, addr & PAGE_MASK, flags);
1da177e4 248
1da177e4 249check_stack:
9b61a4d1
RK
250 /* Don't allow expansion below FIRST_USER_ADDRESS */
251 if (vma->vm_flags & VM_GROWSDOWN &&
252 addr >= FIRST_USER_ADDRESS && !expand_stack(vma, addr))
1da177e4
LT
253 goto good_area;
254out:
255 return fault;
256}
257
785d3cd2 258static int __kprobes
1da177e4
LT
259do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
260{
261 struct task_struct *tsk;
262 struct mm_struct *mm;
2d137c24 263 int fault, sig, code;
8878a539
KC
264 int write = fsr & FSR_WRITE;
265 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
266 (write ? FAULT_FLAG_WRITE : 0);
1da177e4 267
25ce1dd7
NP
268 if (notify_page_fault(regs, fsr))
269 return 0;
270
1da177e4
LT
271 tsk = current;
272 mm = tsk->mm;
273
02fe2845
RK
274 /* Enable interrupts if they were enabled in the parent context. */
275 if (interrupts_enabled(regs))
276 local_irq_enable();
277
1da177e4
LT
278 /*
279 * If we're in an interrupt or have no user
280 * context, we must not take the fault..
281 */
6edaf68a 282 if (in_atomic() || !mm)
1da177e4
LT
283 goto no_context;
284
840ff6a4
RK
285 /*
286 * As per x86, we may deadlock here. However, since the kernel only
287 * validly references user space from well defined areas of the code,
288 * we can bug out early if this is from code which shouldn't.
289 */
290 if (!down_read_trylock(&mm->mmap_sem)) {
291 if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
292 goto no_context;
8878a539 293retry:
840ff6a4 294 down_read(&mm->mmap_sem);
bf456992
RK
295 } else {
296 /*
297 * The above down_read_trylock() might have succeeded in
298 * which case, we'll have missed the might_sleep() from
299 * down_read()
300 */
301 might_sleep();
1d212712
ID
302#ifdef CONFIG_DEBUG_VM
303 if (!user_mode(regs) &&
304 !search_exception_tables(regs->ARM_pc))
305 goto no_context;
306#endif
840ff6a4
RK
307 }
308
8878a539
KC
309 fault = __do_page_fault(mm, addr, fsr, flags, tsk);
310
311 /* If we need to retry but a fatal signal is pending, handle the
312 * signal first. We do not need to release the mmap_sem because
313 * it would already be released in __lock_page_or_retry in
314 * mm/filemap.c. */
315 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
316 return 0;
317
318 /*
319 * Major/minor page fault accounting is only done on the
320 * initial attempt. If we go through a retry, it is extremely
321 * likely that the page will be found in page cache at that point.
322 */
1da177e4 323
a8b0ca17 324 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
dff2aa7a 325 if (!(fault & VM_FAULT_ERROR) && flags & FAULT_FLAG_ALLOW_RETRY) {
8878a539
KC
326 if (fault & VM_FAULT_MAJOR) {
327 tsk->maj_flt++;
328 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
329 regs, addr);
330 } else {
331 tsk->min_flt++;
332 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
333 regs, addr);
334 }
335 if (fault & VM_FAULT_RETRY) {
336 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
337 * of starvation. */
338 flags &= ~FAULT_FLAG_ALLOW_RETRY;
339 goto retry;
340 }
341 }
342
343 up_read(&mm->mmap_sem);
7ada189f 344
1da177e4 345 /*
ff2afb9d 346 * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
1da177e4 347 */
5c72fc5c 348 if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
1da177e4
LT
349 return 0;
350
b42c6344
RK
351 if (fault & VM_FAULT_OOM) {
352 /*
353 * We ran out of memory, call the OOM killer, and return to
354 * userspace (which will retry the fault, or kill us if we
355 * got oom-killed)
356 */
357 pagefault_out_of_memory();
358 return 0;
359 }
360
1da177e4
LT
361 /*
362 * If we are in kernel mode at this point, we
363 * have no context to handle this fault with.
364 */
365 if (!user_mode(regs))
366 goto no_context;
367
83c54070 368 if (fault & VM_FAULT_SIGBUS) {
2d137c24
AM
369 /*
370 * We had some memory, but were unable to
371 * successfully fix up this page fault.
372 */
373 sig = SIGBUS;
374 code = BUS_ADRERR;
83c54070 375 } else {
2d137c24
AM
376 /*
377 * Something tried to access memory that
378 * isn't in our memory map..
379 */
380 sig = SIGSEGV;
381 code = fault == VM_FAULT_BADACCESS ?
382 SEGV_ACCERR : SEGV_MAPERR;
1da177e4 383 }
1da177e4 384
2d137c24
AM
385 __do_user_fault(tsk, addr, fsr, sig, code, regs);
386 return 0;
1da177e4
LT
387
388no_context:
389 __do_kernel_fault(mm, addr, fsr, regs);
390 return 0;
391}
09529f7a
CM
392#else /* CONFIG_MMU */
393static int
394do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
395{
396 return 0;
397}
398#endif /* CONFIG_MMU */
1da177e4
LT
399
400/*
401 * First Level Translation Fault Handler
402 *
403 * We enter here because the first level page table doesn't contain
404 * a valid entry for the address.
405 *
406 * If the address is in kernel space (>= TASK_SIZE), then we are
407 * probably faulting in the vmalloc() area.
408 *
409 * If the init_task's first level page tables contains the relevant
410 * entry, we copy the it to this task. If not, we send the process
411 * a signal, fixup the exception, or oops the kernel.
412 *
413 * NOTE! We MUST NOT take any locks for this case. We may be in an
414 * interrupt or a critical region, and should only copy the information
415 * from the master page table, nothing more.
416 */
09529f7a 417#ifdef CONFIG_MMU
785d3cd2 418static int __kprobes
1da177e4
LT
419do_translation_fault(unsigned long addr, unsigned int fsr,
420 struct pt_regs *regs)
421{
1da177e4
LT
422 unsigned int index;
423 pgd_t *pgd, *pgd_k;
516295e5 424 pud_t *pud, *pud_k;
1da177e4
LT
425 pmd_t *pmd, *pmd_k;
426
427 if (addr < TASK_SIZE)
428 return do_page_fault(addr, fsr, regs);
429
5e27fb78
A
430 if (user_mode(regs))
431 goto bad_area;
432
1da177e4
LT
433 index = pgd_index(addr);
434
1da177e4
LT
435 pgd = cpu_get_pgd() + index;
436 pgd_k = init_mm.pgd + index;
437
438 if (pgd_none(*pgd_k))
439 goto bad_area;
1da177e4
LT
440 if (!pgd_present(*pgd))
441 set_pgd(pgd, *pgd_k);
442
516295e5
RK
443 pud = pud_offset(pgd, addr);
444 pud_k = pud_offset(pgd_k, addr);
445
446 if (pud_none(*pud_k))
447 goto bad_area;
448 if (!pud_present(*pud))
449 set_pud(pud, *pud_k);
450
451 pmd = pmd_offset(pud, addr);
452 pmd_k = pmd_offset(pud_k, addr);
1da177e4 453
f7b8156d
CM
454#ifdef CONFIG_ARM_LPAE
455 /*
456 * Only one hardware entry per PMD with LPAE.
457 */
458 index = 0;
459#else
33a9c41b
KS
460 /*
461 * On ARM one Linux PGD entry contains two hardware entries (see page
462 * tables layout in pgtable.h). We normally guarantee that we always
463 * fill both L1 entries. But create_mapping() doesn't follow the rule.
464 * It can create inidividual L1 entries, so here we have to call
465 * pmd_none() check for the entry really corresponded to address, not
466 * for the first of pair.
467 */
468 index = (addr >> SECTION_SHIFT) & 1;
f7b8156d 469#endif
33a9c41b 470 if (pmd_none(pmd_k[index]))
1da177e4
LT
471 goto bad_area;
472
473 copy_pmd(pmd, pmd_k);
474 return 0;
475
476bad_area:
e5beac37 477 do_bad_area(addr, fsr, regs);
1da177e4
LT
478 return 0;
479}
09529f7a
CM
480#else /* CONFIG_MMU */
481static int
482do_translation_fault(unsigned long addr, unsigned int fsr,
483 struct pt_regs *regs)
484{
485 return 0;
486}
487#endif /* CONFIG_MMU */
1da177e4
LT
488
489/*
490 * Some section permission faults need to be handled gracefully.
491 * They can happen due to a __{get,put}_user during an oops.
492 */
493static int
494do_sect_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
495{
e5beac37 496 do_bad_area(addr, fsr, regs);
1da177e4
LT
497 return 0;
498}
499
500/*
501 * This abort handler always returns "fault".
502 */
503static int
504do_bad(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
505{
506 return 1;
507}
508
136848d4 509struct fsr_info {
1da177e4
LT
510 int (*fn)(unsigned long addr, unsigned int fsr, struct pt_regs *regs);
511 int sig;
cfb0810e 512 int code;
1da177e4 513 const char *name;
1da177e4
LT
514};
515
136848d4 516/* FSR definition */
f7b8156d
CM
517#ifdef CONFIG_ARM_LPAE
518#include "fsr-3level.c"
519#else
136848d4 520#include "fsr-2level.c"
f7b8156d 521#endif
136848d4 522
1da177e4
LT
523void __init
524hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
6338a6aa 525 int sig, int code, const char *name)
1da177e4 526{
6338a6aa
KS
527 if (nr < 0 || nr >= ARRAY_SIZE(fsr_info))
528 BUG();
529
530 fsr_info[nr].fn = fn;
531 fsr_info[nr].sig = sig;
532 fsr_info[nr].code = code;
533 fsr_info[nr].name = name;
1da177e4
LT
534}
535
536/*
537 * Dispatch a data abort to the relevant handler.
538 */
7ab3f8d5 539asmlinkage void __exception
1da177e4
LT
540do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
541{
c88d6aa7 542 const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
cfb0810e 543 struct siginfo info;
1da177e4 544
df297bf6 545 if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
1da177e4
LT
546 return;
547
548 printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
549 inf->name, fsr, addr);
cfb0810e
RK
550
551 info.si_signo = inf->sig;
552 info.si_errno = 0;
553 info.si_code = inf->code;
554 info.si_addr = (void __user *)addr;
1eeb66a1 555 arm_notify_die("", regs, &info, fsr, 0);
1da177e4
LT
556}
557
3a4b5dca
WD
558void __init
559hook_ifault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *),
560 int sig, int code, const char *name)
561{
562 if (nr < 0 || nr >= ARRAY_SIZE(ifsr_info))
563 BUG();
564
565 ifsr_info[nr].fn = fn;
566 ifsr_info[nr].sig = sig;
567 ifsr_info[nr].code = code;
568 ifsr_info[nr].name = name;
569}
570
7ab3f8d5 571asmlinkage void __exception
4fb28474 572do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
1da177e4 573{
d25ef8b8
KS
574 const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
575 struct siginfo info;
576
577 if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
578 return;
579
580 printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
581 inf->name, ifsr, addr);
582
583 info.si_signo = inf->sig;
584 info.si_errno = 0;
585 info.si_code = inf->code;
586 info.si_addr = (void __user *)addr;
587 arm_notify_die("", regs, &info, ifsr, 0);
1da177e4
LT
588}
589
f7b8156d 590#ifndef CONFIG_ARM_LPAE
993bf4ec
KS
591static int __init exceptions_init(void)
592{
593 if (cpu_architecture() >= CPU_ARCH_ARMv6) {
594 hook_fault_code(4, do_translation_fault, SIGSEGV, SEGV_MAPERR,
595 "I-cache maintenance fault");
596 }
597
b8ab5397
KS
598 if (cpu_architecture() >= CPU_ARCH_ARMv7) {
599 /*
600 * TODO: Access flag faults introduced in ARMv6K.
601 * Runtime check for 'K' extension is needed
602 */
603 hook_fault_code(3, do_bad, SIGSEGV, SEGV_MAPERR,
604 "section access flag fault");
605 hook_fault_code(6, do_bad, SIGSEGV, SEGV_MAPERR,
606 "section access flag fault");
607 }
608
993bf4ec
KS
609 return 0;
610}
611
612arch_initcall(exceptions_init);
f7b8156d 613#endif