]> bbs.cooldavid.org Git - net-next-2.6.git/blame - arch/m32r/mm/fault.c
m32r: invoke oom-killer from page fault
[net-next-2.6.git] / arch / m32r / mm / fault.c
CommitLineData
1da177e4
LT
1/*
2 * linux/arch/m32r/mm/fault.c
3 *
4 * Copyright (c) 2001, 2002 Hitoshi Yamamoto, and H. Kondo
5 * Copyright (c) 2004 Naoto Sugai, NIIBE Yutaka
6 *
7 * Some code taken from i386 version.
8 * Copyright (C) 1995 Linus Torvalds
9 */
10
1da177e4
LT
11#include <linux/signal.h>
12#include <linux/sched.h>
13#include <linux/kernel.h>
14#include <linux/errno.h>
15#include <linux/string.h>
16#include <linux/types.h>
17#include <linux/ptrace.h>
18#include <linux/mman.h>
19#include <linux/mm.h>
20#include <linux/smp.h>
1da177e4
LT
21#include <linux/interrupt.h>
22#include <linux/init.h>
23#include <linux/tty.h>
24#include <linux/vt_kern.h> /* For unblank_screen() */
25#include <linux/highmem.h>
26#include <linux/module.h>
27
28#include <asm/m32r.h>
29#include <asm/system.h>
30#include <asm/uaccess.h>
31#include <asm/hardirq.h>
32#include <asm/mmu_context.h>
33#include <asm/tlbflush.h>
34
35extern void die(const char *, struct pt_regs *, long);
36
37#ifndef CONFIG_SMP
38asmlinkage unsigned int tlb_entry_i_dat;
39asmlinkage unsigned int tlb_entry_d_dat;
40#define tlb_entry_i tlb_entry_i_dat
41#define tlb_entry_d tlb_entry_d_dat
42#else
43unsigned int tlb_entry_i_dat[NR_CPUS];
44unsigned int tlb_entry_d_dat[NR_CPUS];
45#define tlb_entry_i tlb_entry_i_dat[smp_processor_id()]
46#define tlb_entry_d tlb_entry_d_dat[smp_processor_id()]
47#endif
48
49extern void init_tlb(void);
50
1da177e4
LT
51/*======================================================================*
52 * do_page_fault()
53 *======================================================================*
54 * This routine handles page faults. It determines the address,
55 * and the problem, and then passes it off to one of the appropriate
56 * routines.
57 *
58 * ARGUMENT:
59 * regs : M32R SP reg.
60 * error_code : See below
61 * address : M32R MMU MDEVA reg. (Operand ACE)
62 * : M32R BPC reg. (Instruction ACE)
63 *
64 * error_code :
65 * bit 0 == 0 means no page found, 1 means protection fault
66 * bit 1 == 0 means read, 1 means write
67 * bit 2 == 0 means kernel, 1 means user-mode
68 * bit 3 == 0 means data, 1 means instruction
69 *======================================================================*/
70#define ACE_PROTECTION 1
71#define ACE_WRITE 2
72#define ACE_USERMODE 4
73#define ACE_INSTRUCTION 8
74
75asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long error_code,
76 unsigned long address)
77{
78 struct task_struct *tsk;
79 struct mm_struct *mm;
80 struct vm_area_struct * vma;
81 unsigned long page, addr;
82 int write;
83c54070 83 int fault;
1da177e4
LT
84 siginfo_t info;
85
86 /*
87 * If BPSW IE bit enable --> set PSW IE bit
88 */
89 if (regs->psw & M32R_PSW_BIE)
90 local_irq_enable();
91
92 tsk = current;
93
94 info.si_code = SEGV_MAPERR;
95
96 /*
97 * We fault-in kernel-space virtual memory on-demand. The
98 * 'reference' page table is init_mm.pgd.
99 *
100 * NOTE! We MUST NOT take any locks for this case. We may
101 * be in an interrupt or a critical region, and should
102 * only copy the information from the master page table,
103 * nothing more.
104 *
105 * This verifies that the fault happens in kernel space
106 * (error_code & ACE_USERMODE) == 0, and that the fault was not a
107 * protection error (error_code & ACE_PROTECTION) == 0.
108 */
109 if (address >= TASK_SIZE && !(error_code & ACE_USERMODE))
110 goto vmalloc_fault;
111
112 mm = tsk->mm;
113
114 /*
115 * If we're in an interrupt or have no user context or are running in an
116 * atomic region then we must not take the fault..
117 */
118 if (in_atomic() || !mm)
119 goto bad_area_nosemaphore;
120
121 /* When running in the kernel we expect faults to occur only to
122 * addresses in user space. All other faults represent errors in the
123 * kernel and should generate an OOPS. Unfortunatly, in the case of an
80f7228b 124 * erroneous fault occurring in a code path which already holds mmap_sem
1da177e4
LT
125 * we will deadlock attempting to validate the fault against the
126 * address space. Luckily the kernel only validly references user
127 * space from well defined areas of code, which are listed in the
128 * exceptions table.
129 *
130 * As the vast majority of faults will be valid we will only perform
131 * the source reference check when there is a possibilty of a deadlock.
132 * Attempt to lock the address space, if we cannot we then validate the
133 * source. If this is invalid we can skip the address space check,
134 * thus avoiding the deadlock.
135 */
136 if (!down_read_trylock(&mm->mmap_sem)) {
137 if ((error_code & ACE_USERMODE) == 0 &&
138 !search_exception_tables(regs->psw))
139 goto bad_area_nosemaphore;
140 down_read(&mm->mmap_sem);
141 }
142
143 vma = find_vma(mm, address);
144 if (!vma)
145 goto bad_area;
146 if (vma->vm_start <= address)
147 goto good_area;
148 if (!(vma->vm_flags & VM_GROWSDOWN))
149 goto bad_area;
6b8bd3f4 150
1da177e4
LT
151 if (error_code & ACE_USERMODE) {
152 /*
153 * accessing the stack below "spu" is always a bug.
154 * The "+ 4" is there due to the push instruction
155 * doing pre-decrement on the stack and that
156 * doesn't show up until later..
157 */
158 if (address + 4 < regs->spu)
159 goto bad_area;
160 }
6b8bd3f4 161
1da177e4
LT
162 if (expand_stack(vma, address))
163 goto bad_area;
164/*
165 * Ok, we have a good vm_area for this memory access, so
166 * we can handle it..
167 */
168good_area:
169 info.si_code = SEGV_ACCERR;
170 write = 0;
171 switch (error_code & (ACE_WRITE|ACE_PROTECTION)) {
172 default: /* 3: write, present */
173 /* fall through */
174 case ACE_WRITE: /* write, not present */
175 if (!(vma->vm_flags & VM_WRITE))
176 goto bad_area;
177 write++;
178 break;
179 case ACE_PROTECTION: /* read, present */
180 case 0: /* read, not present */
181 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
182 goto bad_area;
183 }
184
185 /*
186 * For instruction access exception, check if the area is executable
187 */
188 if ((error_code & ACE_INSTRUCTION) && !(vma->vm_flags & VM_EXEC))
189 goto bad_area;
190
1da177e4
LT
191 /*
192 * If for any reason at all we couldn't handle the fault,
193 * make sure we exit gracefully rather than endlessly redo
194 * the fault.
195 */
196 addr = (address & PAGE_MASK);
197 set_thread_fault_code(error_code);
d06063cc 198 fault = handle_mm_fault(mm, vma, addr, write ? FAULT_FLAG_WRITE : 0);
83c54070
NP
199 if (unlikely(fault & VM_FAULT_ERROR)) {
200 if (fault & VM_FAULT_OOM)
1da177e4 201 goto out_of_memory;
83c54070
NP
202 else if (fault & VM_FAULT_SIGBUS)
203 goto do_sigbus;
204 BUG();
1da177e4 205 }
83c54070
NP
206 if (fault & VM_FAULT_MAJOR)
207 tsk->maj_flt++;
208 else
209 tsk->min_flt++;
1da177e4
LT
210 set_thread_fault_code(0);
211 up_read(&mm->mmap_sem);
212 return;
213
214/*
215 * Something tried to access memory that isn't in our memory map..
216 * Fix it, but check if it's kernel or user first..
217 */
218bad_area:
219 up_read(&mm->mmap_sem);
220
221bad_area_nosemaphore:
222 /* User mode accesses just cause a SIGSEGV */
223 if (error_code & ACE_USERMODE) {
224 tsk->thread.address = address;
225 tsk->thread.error_code = error_code | (address >= TASK_SIZE);
226 tsk->thread.trap_no = 14;
227 info.si_signo = SIGSEGV;
228 info.si_errno = 0;
229 /* info.si_code has been set above */
230 info.si_addr = (void __user *)address;
231 force_sig_info(SIGSEGV, &info, tsk);
232 return;
233 }
234
235no_context:
236 /* Are we prepared to handle this kernel fault? */
237 if (fixup_exception(regs))
238 return;
239
240/*
241 * Oops. The kernel tried to access some bad page. We'll have to
242 * terminate things with extreme prejudice.
243 */
244
245 bust_spinlocks(1);
246
247 if (address < PAGE_SIZE)
248 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference");
249 else
250 printk(KERN_ALERT "Unable to handle kernel paging request");
251 printk(" at virtual address %08lx\n",address);
252 printk(KERN_ALERT " printing bpc:\n");
253 printk("%08lx\n", regs->bpc);
254 page = *(unsigned long *)MPTB;
255 page = ((unsigned long *) page)[address >> PGDIR_SHIFT];
256 printk(KERN_ALERT "*pde = %08lx\n", page);
257 if (page & _PAGE_PRESENT) {
258 page &= PAGE_MASK;
259 address &= 0x003ff000;
260 page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
261 printk(KERN_ALERT "*pte = %08lx\n", page);
262 }
263 die("Oops", regs, error_code);
264 bust_spinlocks(0);
265 do_exit(SIGKILL);
266
267/*
268 * We ran out of memory, or some other thing happened to us that made
269 * us unable to handle the page fault gracefully.
270 */
271out_of_memory:
272 up_read(&mm->mmap_sem);
68db30ce
NP
273 if (!(error_code & ACE_USERMODE))
274 goto no_context;
275 pagefault_out_of_memory();
276 return;
1da177e4
LT
277
278do_sigbus:
279 up_read(&mm->mmap_sem);
280
281 /* Kernel mode? Handle exception or die */
282 if (!(error_code & ACE_USERMODE))
283 goto no_context;
284
285 tsk->thread.address = address;
286 tsk->thread.error_code = error_code;
287 tsk->thread.trap_no = 14;
288 info.si_signo = SIGBUS;
289 info.si_errno = 0;
290 info.si_code = BUS_ADRERR;
291 info.si_addr = (void __user *)address;
292 force_sig_info(SIGBUS, &info, tsk);
293 return;
294
295vmalloc_fault:
296 {
297 /*
298 * Synchronize this task's top level page-table
299 * with the 'reference' page table.
300 *
301 * Do _not_ use "tsk" here. We might be inside
302 * an interrupt in the middle of a task switch..
303 */
304 int offset = pgd_index(address);
305 pgd_t *pgd, *pgd_k;
306 pmd_t *pmd, *pmd_k;
307 pte_t *pte_k;
308
309 pgd = (pgd_t *)*(unsigned long *)MPTB;
310 pgd = offset + (pgd_t *)pgd;
311 pgd_k = init_mm.pgd + offset;
312
313 if (!pgd_present(*pgd_k))
314 goto no_context;
315
316 /*
317 * set_pgd(pgd, *pgd_k); here would be useless on PAE
318 * and redundant with the set_pmd() on non-PAE.
319 */
320
321 pmd = pmd_offset(pgd, address);
322 pmd_k = pmd_offset(pgd_k, address);
323 if (!pmd_present(*pmd_k))
324 goto no_context;
325 set_pmd(pmd, *pmd_k);
326
327 pte_k = pte_offset_kernel(pmd_k, address);
328 if (!pte_present(*pte_k))
329 goto no_context;
330
9b87ed79
HT
331 addr = (address & PAGE_MASK);
332 set_thread_fault_code(error_code);
4b3073e1 333 update_mmu_cache(NULL, addr, pte_k);
9b87ed79 334 set_thread_fault_code(0);
1da177e4
LT
335 return;
336 }
337}
338
339/*======================================================================*
340 * update_mmu_cache()
341 *======================================================================*/
342#define TLB_MASK (NR_TLB_ENTRIES - 1)
343#define ITLB_END (unsigned long *)(ITLB_BASE + (NR_TLB_ENTRIES * 8))
344#define DTLB_END (unsigned long *)(DTLB_BASE + (NR_TLB_ENTRIES * 8))
345void update_mmu_cache(struct vm_area_struct *vma, unsigned long vaddr,
4b3073e1 346 pte_t *ptep)
1da177e4 347{
9b87ed79 348 volatile unsigned long *entry1, *entry2;
1da177e4
LT
349 unsigned long pte_data, flags;
350 unsigned int *entry_dat;
351 int inst = get_thread_fault_code() & ACE_INSTRUCTION;
352 int i;
353
354 /* Ptrace may call this routine. */
355 if (vma && current->active_mm != vma->vm_mm)
356 return;
357
358 local_irq_save(flags);
359
360 vaddr = (vaddr & PAGE_MASK) | get_asid();
361
4b3073e1 362 pte_data = pte_val(*ptep);
9b87ed79 363
1da177e4
LT
364#ifdef CONFIG_CHIP_OPSP
365 entry1 = (unsigned long *)ITLB_BASE;
9b87ed79
HT
366 for (i = 0; i < NR_TLB_ENTRIES; i++) {
367 if (*entry1++ == vaddr) {
368 set_tlb_data(entry1, pte_data);
369 break;
370 }
371 entry1++;
1da177e4
LT
372 }
373 entry2 = (unsigned long *)DTLB_BASE;
9b87ed79
HT
374 for (i = 0; i < NR_TLB_ENTRIES; i++) {
375 if (*entry2++ == vaddr) {
376 set_tlb_data(entry2, pte_data);
377 break;
378 }
379 entry2++;
1da177e4 380 }
1da177e4 381#else
1da177e4
LT
382 /*
383 * Update TLB entries
384 * entry1: ITLB entry address
385 * entry2: DTLB entry address
386 */
387 __asm__ __volatile__ (
388 "seth %0, #high(%4) \n\t"
389 "st %2, @(%5, %0) \n\t"
390 "ldi %1, #1 \n\t"
391 "st %1, @(%6, %0) \n\t"
392 "add3 r4, %0, %7 \n\t"
393 ".fillinsn \n"
394 "1: \n\t"
395 "ld %1, @(%6, %0) \n\t"
396 "bnez %1, 1b \n\t"
397 "ld %0, @r4+ \n\t"
398 "ld %1, @r4 \n\t"
399 "st %3, @+%0 \n\t"
400 "st %3, @+%1 \n\t"
401 : "=&r" (entry1), "=&r" (entry2)
402 : "r" (vaddr), "r" (pte_data), "i" (MMU_REG_BASE),
403 "i" (MSVA_offset), "i" (MTOP_offset), "i" (MIDXI_offset)
404 : "r4", "memory"
405 );
9b87ed79 406#endif
1da177e4
LT
407
408 if ((!inst && entry2 >= DTLB_END) || (inst && entry1 >= ITLB_END))
409 goto notfound;
410
411found:
412 local_irq_restore(flags);
413
414 return;
415
416 /* Valid entry not found */
417notfound:
418 /*
419 * Update ITLB or DTLB entry
420 * entry1: TLB entry address
421 * entry2: TLB base address
422 */
423 if (!inst) {
424 entry2 = (unsigned long *)DTLB_BASE;
425 entry_dat = &tlb_entry_d;
426 } else {
427 entry2 = (unsigned long *)ITLB_BASE;
428 entry_dat = &tlb_entry_i;
429 }
430 entry1 = entry2 + (((*entry_dat - 1) & TLB_MASK) << 1);
431
432 for (i = 0 ; i < NR_TLB_ENTRIES ; i++) {
433 if (!(entry1[1] & 2)) /* Valid bit check */
434 break;
435
436 if (entry1 != entry2)
437 entry1 -= 2;
438 else
439 entry1 += TLB_MASK << 1;
440 }
441
442 if (i >= NR_TLB_ENTRIES) { /* Empty entry not found */
443 entry1 = entry2 + (*entry_dat << 1);
444 *entry_dat = (*entry_dat + 1) & TLB_MASK;
445 }
446 *entry1++ = vaddr; /* Set TLB tag */
447 set_tlb_data(entry1, pte_data);
448
449 goto found;
1da177e4
LT
450}
451
452/*======================================================================*
453 * flush_tlb_page() : flushes one page
454 *======================================================================*/
455void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
456{
457 if (vma->vm_mm && mm_context(vma->vm_mm) != NO_CONTEXT) {
458 unsigned long flags;
459
460 local_irq_save(flags);
461 page &= PAGE_MASK;
462 page |= (mm_context(vma->vm_mm) & MMU_CONTEXT_ASID_MASK);
463 __flush_tlb_page(page);
464 local_irq_restore(flags);
465 }
466}
467
468/*======================================================================*
469 * flush_tlb_range() : flushes a range of pages
470 *======================================================================*/
471void local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
472 unsigned long end)
473{
474 struct mm_struct *mm;
475
476 mm = vma->vm_mm;
477 if (mm_context(mm) != NO_CONTEXT) {
478 unsigned long flags;
479 int size;
480
481 local_irq_save(flags);
482 size = (end - start + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
483 if (size > (NR_TLB_ENTRIES / 4)) { /* Too many TLB to flush */
484 mm_context(mm) = NO_CONTEXT;
485 if (mm == current->mm)
486 activate_context(mm);
487 } else {
488 unsigned long asid;
489
490 asid = mm_context(mm) & MMU_CONTEXT_ASID_MASK;
491 start &= PAGE_MASK;
492 end += (PAGE_SIZE - 1);
493 end &= PAGE_MASK;
494
495 start |= asid;
496 end |= asid;
497 while (start < end) {
498 __flush_tlb_page(start);
499 start += PAGE_SIZE;
500 }
501 }
502 local_irq_restore(flags);
503 }
504}
505
506/*======================================================================*
507 * flush_tlb_mm() : flushes the specified mm context TLB's
508 *======================================================================*/
509void local_flush_tlb_mm(struct mm_struct *mm)
510{
511 /* Invalidate all TLB of this process. */
512 /* Instead of invalidating each TLB, we get new MMU context. */
513 if (mm_context(mm) != NO_CONTEXT) {
514 unsigned long flags;
515
516 local_irq_save(flags);
517 mm_context(mm) = NO_CONTEXT;
518 if (mm == current->mm)
519 activate_context(mm);
520 local_irq_restore(flags);
521 }
522}
523
524/*======================================================================*
525 * flush_tlb_all() : flushes all processes TLBs
526 *======================================================================*/
527void local_flush_tlb_all(void)
528{
529 unsigned long flags;
530
531 local_irq_save(flags);
532 __flush_tlb_all();
533 local_irq_restore(flags);
534}
535
536/*======================================================================*
537 * init_mmu()
538 *======================================================================*/
539void __init init_mmu(void)
540{
541 tlb_entry_i = 0;
542 tlb_entry_d = 0;
543 mmu_context_cache = MMU_CONTEXT_FIRST_VERSION;
544 set_asid(mmu_context_cache & MMU_CONTEXT_ASID_MASK);
545 *(volatile unsigned long *)MPTB = (unsigned long)swapper_pg_dir;
546}