Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3eeffb32 | 2 | * arch/sh/mm/tlb-flush_64.c |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 2000, 2001 Paolo Alberelli | |
5 | * Copyright (C) 2003 Richard Curnow (/proc/tlb, bug fixes) | |
6 | * Copyright (C) 2003 Paul Mundt | |
7 | * | |
3eeffb32 PM |
8 | * This file is subject to the terms and conditions of the GNU General Public |
9 | * License. See the file "COPYING" in the main directory of this archive | |
10 | * for more details. | |
1da177e4 | 11 | */ |
1da177e4 LT |
12 | #include <linux/signal.h> |
13 | #include <linux/rwsem.h> | |
14 | #include <linux/sched.h> | |
15 | #include <linux/kernel.h> | |
16 | #include <linux/errno.h> | |
17 | #include <linux/string.h> | |
18 | #include <linux/types.h> | |
19 | #include <linux/ptrace.h> | |
20 | #include <linux/mman.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/smp.h> | |
1da177e4 | 23 | #include <linux/interrupt.h> |
1da177e4 LT |
24 | #include <asm/system.h> |
25 | #include <asm/io.h> | |
26 | #include <asm/tlb.h> | |
27 | #include <asm/uaccess.h> | |
28 | #include <asm/pgalloc.h> | |
29 | #include <asm/mmu_context.h> | |
1da177e4 LT |
30 | |
31 | extern void die(const char *,struct pt_regs *,long); | |
32 | ||
33 | #define PFLAG(val,flag) (( (val) & (flag) ) ? #flag : "" ) | |
34 | #define PPROT(flag) PFLAG(pgprot_val(prot),flag) | |
35 | ||
36 | static inline void print_prots(pgprot_t prot) | |
37 | { | |
38 | printk("prot is 0x%08lx\n",pgprot_val(prot)); | |
39 | ||
40 | printk("%s %s %s %s %s\n",PPROT(_PAGE_SHARED),PPROT(_PAGE_READ), | |
41 | PPROT(_PAGE_EXECUTE),PPROT(_PAGE_WRITE),PPROT(_PAGE_USER)); | |
42 | } | |
43 | ||
44 | static inline void print_vma(struct vm_area_struct *vma) | |
45 | { | |
46 | printk("vma start 0x%08lx\n", vma->vm_start); | |
47 | printk("vma end 0x%08lx\n", vma->vm_end); | |
48 | ||
49 | print_prots(vma->vm_page_prot); | |
50 | printk("vm_flags 0x%08lx\n", vma->vm_flags); | |
51 | } | |
52 | ||
53 | static inline void print_task(struct task_struct *tsk) | |
54 | { | |
19c5870c | 55 | printk("Task pid %d\n", task_pid_nr(tsk)); |
1da177e4 LT |
56 | } |
57 | ||
58 | static pte_t *lookup_pte(struct mm_struct *mm, unsigned long address) | |
59 | { | |
60 | pgd_t *dir; | |
3eeffb32 | 61 | pud_t *pud; |
1da177e4 LT |
62 | pmd_t *pmd; |
63 | pte_t *pte; | |
64 | pte_t entry; | |
65 | ||
66 | dir = pgd_offset(mm, address); | |
3eeffb32 | 67 | if (pgd_none(*dir)) |
1da177e4 | 68 | return NULL; |
1da177e4 | 69 | |
3eeffb32 PM |
70 | pud = pud_offset(dir, address); |
71 | if (pud_none(*pud)) | |
72 | return NULL; | |
73 | ||
74 | pmd = pmd_offset(pud, address); | |
75 | if (pmd_none(*pmd)) | |
1da177e4 | 76 | return NULL; |
1da177e4 LT |
77 | |
78 | pte = pte_offset_kernel(pmd, address); | |
79 | entry = *pte; | |
3eeffb32 | 80 | if (pte_none(entry) || !pte_present(entry)) |
1da177e4 | 81 | return NULL; |
1da177e4 LT |
82 | |
83 | return pte; | |
84 | } | |
85 | ||
86 | /* | |
87 | * This routine handles page faults. It determines the address, | |
88 | * and the problem, and then passes it off to one of the appropriate | |
89 | * routines. | |
90 | */ | |
91 | asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long writeaccess, | |
92 | unsigned long textaccess, unsigned long address) | |
93 | { | |
94 | struct task_struct *tsk; | |
95 | struct mm_struct *mm; | |
96 | struct vm_area_struct * vma; | |
97 | const struct exception_table_entry *fixup; | |
98 | pte_t *pte; | |
83c54070 | 99 | int fault; |
1da177e4 | 100 | |
1da177e4 LT |
101 | /* SIM |
102 | * Note this is now called with interrupts still disabled | |
103 | * This is to cope with being called for a missing IO port | |
0a354775 | 104 | * address with interrupts disabled. This should be fixed as |
1da177e4 LT |
105 | * soon as we have a better 'fast path' miss handler. |
106 | * | |
107 | * Plus take care how you try and debug this stuff. | |
108 | * For example, writing debug data to a port which you | |
109 | * have just faulted on is not going to work. | |
110 | */ | |
111 | ||
112 | tsk = current; | |
113 | mm = tsk->mm; | |
114 | ||
115 | /* Not an IO address, so reenable interrupts */ | |
116 | local_irq_enable(); | |
117 | ||
118 | /* | |
119 | * If we're in an interrupt or have no user | |
120 | * context, we must not take the fault.. | |
121 | */ | |
6edaf68a | 122 | if (in_atomic() || !mm) |
1da177e4 LT |
123 | goto no_context; |
124 | ||
125 | /* TLB misses upon some cache flushes get done under cli() */ | |
126 | down_read(&mm->mmap_sem); | |
127 | ||
128 | vma = find_vma(mm, address); | |
129 | ||
130 | if (!vma) { | |
131 | #ifdef DEBUG_FAULT | |
132 | print_task(tsk); | |
133 | printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n", | |
866e6b9e | 134 | __func__, __LINE__, |
1da177e4 LT |
135 | address,regs->pc,textaccess,writeaccess); |
136 | show_regs(regs); | |
137 | #endif | |
138 | goto bad_area; | |
139 | } | |
140 | if (vma->vm_start <= address) { | |
141 | goto good_area; | |
142 | } | |
143 | ||
144 | if (!(vma->vm_flags & VM_GROWSDOWN)) { | |
145 | #ifdef DEBUG_FAULT | |
146 | print_task(tsk); | |
147 | printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n", | |
866e6b9e | 148 | __func__, __LINE__, |
1da177e4 LT |
149 | address,regs->pc,textaccess,writeaccess); |
150 | show_regs(regs); | |
151 | ||
152 | print_vma(vma); | |
153 | #endif | |
154 | goto bad_area; | |
155 | } | |
156 | if (expand_stack(vma, address)) { | |
157 | #ifdef DEBUG_FAULT | |
158 | print_task(tsk); | |
159 | printk("%s:%d fault, address is 0x%08x PC %016Lx textaccess %d writeaccess %d\n", | |
866e6b9e | 160 | __func__, __LINE__, |
1da177e4 LT |
161 | address,regs->pc,textaccess,writeaccess); |
162 | show_regs(regs); | |
163 | #endif | |
164 | goto bad_area; | |
165 | } | |
166 | /* | |
167 | * Ok, we have a good vm_area for this memory access, so | |
168 | * we can handle it.. | |
169 | */ | |
170 | good_area: | |
171 | if (textaccess) { | |
172 | if (!(vma->vm_flags & VM_EXEC)) | |
173 | goto bad_area; | |
174 | } else { | |
175 | if (writeaccess) { | |
176 | if (!(vma->vm_flags & VM_WRITE)) | |
177 | goto bad_area; | |
178 | } else { | |
179 | if (!(vma->vm_flags & VM_READ)) | |
180 | goto bad_area; | |
181 | } | |
182 | } | |
183 | ||
184 | /* | |
185 | * If for any reason at all we couldn't handle the fault, | |
186 | * make sure we exit gracefully rather than endlessly redo | |
187 | * the fault. | |
188 | */ | |
189 | survive: | |
83c54070 NP |
190 | fault = handle_mm_fault(mm, vma, address, writeaccess); |
191 | if (unlikely(fault & VM_FAULT_ERROR)) { | |
192 | if (fault & VM_FAULT_OOM) | |
193 | goto out_of_memory; | |
194 | else if (fault & VM_FAULT_SIGBUS) | |
195 | goto do_sigbus; | |
196 | BUG(); | |
1da177e4 | 197 | } |
83c54070 NP |
198 | if (fault & VM_FAULT_MAJOR) |
199 | tsk->maj_flt++; | |
200 | else | |
201 | tsk->min_flt++; | |
202 | ||
1da177e4 LT |
203 | /* If we get here, the page fault has been handled. Do the TLB refill |
204 | now from the newly-setup PTE, to avoid having to fault again right | |
205 | away on the same instruction. */ | |
206 | pte = lookup_pte (mm, address); | |
207 | if (!pte) { | |
208 | /* From empirical evidence, we can get here, due to | |
209 | !pte_present(pte). (e.g. if a swap-in occurs, and the page | |
210 | is swapped back out again before the process that wanted it | |
211 | gets rescheduled?) */ | |
212 | goto no_pte; | |
213 | } | |
214 | ||
215 | __do_tlb_refill(address, textaccess, pte); | |
216 | ||
217 | no_pte: | |
218 | ||
219 | up_read(&mm->mmap_sem); | |
220 | return; | |
221 | ||
222 | /* | |
223 | * Something tried to access memory that isn't in our memory map.. | |
224 | * Fix it, but check if it's kernel or user first.. | |
225 | */ | |
226 | bad_area: | |
227 | #ifdef DEBUG_FAULT | |
228 | printk("fault:bad area\n"); | |
229 | #endif | |
230 | up_read(&mm->mmap_sem); | |
231 | ||
232 | if (user_mode(regs)) { | |
233 | static int count=0; | |
234 | siginfo_t info; | |
235 | if (count < 4) { | |
236 | /* This is really to help debug faults when starting | |
237 | * usermode, so only need a few */ | |
238 | count++; | |
239 | printk("user mode bad_area address=%08lx pid=%d (%s) pc=%08lx\n", | |
19c5870c | 240 | address, task_pid_nr(current), current->comm, |
1da177e4 LT |
241 | (unsigned long) regs->pc); |
242 | #if 0 | |
243 | show_regs(regs); | |
244 | #endif | |
245 | } | |
b460cbc5 | 246 | if (is_global_init(tsk)) { |
1da177e4 LT |
247 | panic("INIT had user mode bad_area\n"); |
248 | } | |
249 | tsk->thread.address = address; | |
250 | tsk->thread.error_code = writeaccess; | |
251 | info.si_signo = SIGSEGV; | |
252 | info.si_errno = 0; | |
253 | info.si_addr = (void *) address; | |
254 | force_sig_info(SIGSEGV, &info, tsk); | |
255 | return; | |
256 | } | |
257 | ||
258 | no_context: | |
259 | #ifdef DEBUG_FAULT | |
260 | printk("fault:No context\n"); | |
261 | #endif | |
262 | /* Are we prepared to handle this kernel fault? */ | |
263 | fixup = search_exception_tables(regs->pc); | |
264 | if (fixup) { | |
265 | regs->pc = fixup->fixup; | |
266 | return; | |
267 | } | |
268 | ||
269 | /* | |
270 | * Oops. The kernel tried to access some bad page. We'll have to | |
271 | * terminate things with extreme prejudice. | |
272 | * | |
273 | */ | |
274 | if (address < PAGE_SIZE) | |
275 | printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference"); | |
276 | else | |
277 | printk(KERN_ALERT "Unable to handle kernel paging request"); | |
278 | printk(" at virtual address %08lx\n", address); | |
279 | printk(KERN_ALERT "pc = %08Lx%08Lx\n", regs->pc >> 32, regs->pc & 0xffffffff); | |
280 | die("Oops", regs, writeaccess); | |
281 | do_exit(SIGKILL); | |
282 | ||
283 | /* | |
284 | * We ran out of memory, or some other thing happened to us that made | |
285 | * us unable to handle the page fault gracefully. | |
286 | */ | |
287 | out_of_memory: | |
b460cbc5 | 288 | if (is_global_init(current)) { |
1da177e4 LT |
289 | panic("INIT out of memory\n"); |
290 | yield(); | |
291 | goto survive; | |
292 | } | |
293 | printk("fault:Out of memory\n"); | |
294 | up_read(&mm->mmap_sem); | |
b460cbc5 | 295 | if (is_global_init(current)) { |
1da177e4 LT |
296 | yield(); |
297 | down_read(&mm->mmap_sem); | |
298 | goto survive; | |
299 | } | |
300 | printk("VM: killing process %s\n", tsk->comm); | |
301 | if (user_mode(regs)) | |
dcca2bde | 302 | do_group_exit(SIGKILL); |
1da177e4 LT |
303 | goto no_context; |
304 | ||
305 | do_sigbus: | |
306 | printk("fault:Do sigbus\n"); | |
307 | up_read(&mm->mmap_sem); | |
308 | ||
309 | /* | |
310 | * Send a sigbus, regardless of whether we were in kernel | |
311 | * or user mode. | |
312 | */ | |
313 | tsk->thread.address = address; | |
314 | tsk->thread.error_code = writeaccess; | |
315 | tsk->thread.trap_no = 14; | |
316 | force_sig(SIGBUS, tsk); | |
317 | ||
318 | /* Kernel mode? Handle exceptions or die */ | |
319 | if (!user_mode(regs)) | |
320 | goto no_context; | |
321 | } | |
322 | ||
1da177e4 LT |
323 | void update_mmu_cache(struct vm_area_struct * vma, |
324 | unsigned long address, pte_t pte) | |
325 | { | |
1da177e4 LT |
326 | /* |
327 | * This appears to get called once for every pte entry that gets | |
328 | * established => I don't think it's efficient to try refilling the | |
329 | * TLBs with the pages - some may not get accessed even. Also, for | |
330 | * executable pages, it is impossible to determine reliably here which | |
331 | * TLB they should be mapped into (or both even). | |
332 | * | |
333 | * So, just do nothing here and handle faults on demand. In the | |
334 | * TLBMISS handling case, the refill is now done anyway after the pte | |
335 | * has been fixed up, so that deals with most useful cases. | |
336 | */ | |
337 | } | |
338 | ||
3eeffb32 | 339 | void local_flush_tlb_one(unsigned long asid, unsigned long page) |
1da177e4 LT |
340 | { |
341 | unsigned long long match, pteh=0, lpage; | |
342 | unsigned long tlb; | |
1da177e4 LT |
343 | |
344 | /* | |
345 | * Sign-extend based on neff. | |
346 | */ | |
347 | lpage = (page & NEFF_SIGN) ? (page | NEFF_MASK) : page; | |
3eeffb32 | 348 | match = (asid << PTEH_ASID_SHIFT) | PTEH_VALID; |
1da177e4 LT |
349 | match |= lpage; |
350 | ||
3eeffb32 PM |
351 | for_each_itlb_entry(tlb) { |
352 | asm volatile ("getcfg %1, 0, %0" | |
353 | : "=r" (pteh) | |
354 | : "r" (tlb) ); | |
1da177e4 | 355 | |
3eeffb32 PM |
356 | if (pteh == match) { |
357 | __flush_tlb_slot(tlb); | |
358 | break; | |
1da177e4 LT |
359 | } |
360 | } | |
361 | ||
1da177e4 LT |
362 | for_each_dtlb_entry(tlb) { |
363 | asm volatile ("getcfg %1, 0, %0" | |
364 | : "=r" (pteh) | |
365 | : "r" (tlb) ); | |
366 | ||
367 | if (pteh == match) { | |
368 | __flush_tlb_slot(tlb); | |
369 | break; | |
370 | } | |
371 | ||
372 | } | |
373 | } | |
374 | ||
3eeffb32 | 375 | void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long page) |
1da177e4 LT |
376 | { |
377 | unsigned long flags; | |
378 | ||
1da177e4 LT |
379 | if (vma->vm_mm) { |
380 | page &= PAGE_MASK; | |
381 | local_irq_save(flags); | |
3eeffb32 | 382 | local_flush_tlb_one(get_asid(), page); |
1da177e4 LT |
383 | local_irq_restore(flags); |
384 | } | |
385 | } | |
386 | ||
3eeffb32 PM |
387 | void local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start, |
388 | unsigned long end) | |
1da177e4 LT |
389 | { |
390 | unsigned long flags; | |
391 | unsigned long long match, pteh=0, pteh_epn, pteh_low; | |
392 | unsigned long tlb; | |
3eeffb32 | 393 | unsigned int cpu = smp_processor_id(); |
1da177e4 LT |
394 | struct mm_struct *mm; |
395 | ||
396 | mm = vma->vm_mm; | |
3eeffb32 | 397 | if (cpu_context(cpu, mm) == NO_CONTEXT) |
1da177e4 LT |
398 | return; |
399 | ||
400 | local_irq_save(flags); | |
401 | ||
402 | start &= PAGE_MASK; | |
403 | end &= PAGE_MASK; | |
404 | ||
3eeffb32 | 405 | match = (cpu_asid(cpu, mm) << PTEH_ASID_SHIFT) | PTEH_VALID; |
1da177e4 LT |
406 | |
407 | /* Flush ITLB */ | |
408 | for_each_itlb_entry(tlb) { | |
409 | asm volatile ("getcfg %1, 0, %0" | |
410 | : "=r" (pteh) | |
411 | : "r" (tlb) ); | |
412 | ||
413 | pteh_epn = pteh & PAGE_MASK; | |
414 | pteh_low = pteh & ~PAGE_MASK; | |
415 | ||
416 | if (pteh_low == match && pteh_epn >= start && pteh_epn <= end) | |
417 | __flush_tlb_slot(tlb); | |
418 | } | |
419 | ||
420 | /* Flush DTLB */ | |
421 | for_each_dtlb_entry(tlb) { | |
422 | asm volatile ("getcfg %1, 0, %0" | |
423 | : "=r" (pteh) | |
424 | : "r" (tlb) ); | |
425 | ||
426 | pteh_epn = pteh & PAGE_MASK; | |
427 | pteh_low = pteh & ~PAGE_MASK; | |
428 | ||
429 | if (pteh_low == match && pteh_epn >= start && pteh_epn <= end) | |
430 | __flush_tlb_slot(tlb); | |
431 | } | |
432 | ||
433 | local_irq_restore(flags); | |
434 | } | |
435 | ||
3eeffb32 | 436 | void local_flush_tlb_mm(struct mm_struct *mm) |
1da177e4 LT |
437 | { |
438 | unsigned long flags; | |
3eeffb32 | 439 | unsigned int cpu = smp_processor_id(); |
1da177e4 | 440 | |
3eeffb32 | 441 | if (cpu_context(cpu, mm) == NO_CONTEXT) |
1da177e4 LT |
442 | return; |
443 | ||
444 | local_irq_save(flags); | |
445 | ||
3eeffb32 PM |
446 | cpu_context(cpu, mm) = NO_CONTEXT; |
447 | if (mm == current->mm) | |
448 | activate_context(mm, cpu); | |
1da177e4 LT |
449 | |
450 | local_irq_restore(flags); | |
1da177e4 LT |
451 | } |
452 | ||
3eeffb32 | 453 | void local_flush_tlb_all(void) |
1da177e4 LT |
454 | { |
455 | /* Invalidate all, including shared pages, excluding fixed TLBs */ | |
1da177e4 LT |
456 | unsigned long flags, tlb; |
457 | ||
1da177e4 LT |
458 | local_irq_save(flags); |
459 | ||
460 | /* Flush each ITLB entry */ | |
3eeffb32 | 461 | for_each_itlb_entry(tlb) |
1da177e4 | 462 | __flush_tlb_slot(tlb); |
1da177e4 LT |
463 | |
464 | /* Flush each DTLB entry */ | |
3eeffb32 | 465 | for_each_dtlb_entry(tlb) |
1da177e4 | 466 | __flush_tlb_slot(tlb); |
1da177e4 LT |
467 | |
468 | local_irq_restore(flags); | |
469 | } | |
470 | ||
3eeffb32 | 471 | void local_flush_tlb_kernel_range(unsigned long start, unsigned long end) |
1da177e4 LT |
472 | { |
473 | /* FIXME: Optimize this later.. */ | |
474 | flush_tlb_all(); | |
475 | } |