perf: Remove the nmi parameter from the swevent and overflow interface
[linux-2.6.git] / arch / sparc / mm / fault_32.c
1 /*
2  * fault.c:  Page fault handlers for the Sparc.
3  *
4  * Copyright (C) 1995 David S. Miller (davem@caip.rutgers.edu)
5  * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be)
6  * Copyright (C) 1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
7  */
8
9 #include <asm/head.h>
10
11 #include <linux/string.h>
12 #include <linux/types.h>
13 #include <linux/sched.h>
14 #include <linux/ptrace.h>
15 #include <linux/mman.h>
16 #include <linux/threads.h>
17 #include <linux/kernel.h>
18 #include <linux/signal.h>
19 #include <linux/mm.h>
20 #include <linux/smp.h>
21 #include <linux/perf_event.h>
22 #include <linux/interrupt.h>
23 #include <linux/module.h>
24 #include <linux/kdebug.h>
25
26 #include <asm/system.h>
27 #include <asm/page.h>
28 #include <asm/pgtable.h>
29 #include <asm/memreg.h>
30 #include <asm/openprom.h>
31 #include <asm/oplib.h>
32 #include <asm/smp.h>
33 #include <asm/traps.h>
34 #include <asm/uaccess.h>
35
36 extern int prom_node_root;
37
38 int show_unhandled_signals = 1;
39
40 /* At boot time we determine these two values necessary for setting
41  * up the segment maps and page table entries (pte's).
42  */
43
44 int num_segmaps, num_contexts;
45 int invalid_segment;
46
47 /* various Virtual Address Cache parameters we find at boot time... */
48
49 int vac_size, vac_linesize, vac_do_hw_vac_flushes;
50 int vac_entries_per_context, vac_entries_per_segment;
51 int vac_entries_per_page;
52
53 /* Return how much physical memory we have.  */
54 unsigned long probe_memory(void)
55 {
56         unsigned long total = 0;
57         int i;
58
59         for (i = 0; sp_banks[i].num_bytes; i++)
60                 total += sp_banks[i].num_bytes;
61
62         return total;
63 }
64
65 extern void sun4c_complete_all_stores(void);
66
67 /* Whee, a level 15 NMI interrupt memory error.  Let's have fun... */
68 asmlinkage void sparc_lvl15_nmi(struct pt_regs *regs, unsigned long serr,
69                                 unsigned long svaddr, unsigned long aerr,
70                                 unsigned long avaddr)
71 {
72         sun4c_complete_all_stores();
73         printk("FAULT: NMI received\n");
74         printk("SREGS: Synchronous Error %08lx\n", serr);
75         printk("       Synchronous Vaddr %08lx\n", svaddr);
76         printk("      Asynchronous Error %08lx\n", aerr);
77         printk("      Asynchronous Vaddr %08lx\n", avaddr);
78         if (sun4c_memerr_reg)
79                 printk("     Memory Parity Error %08lx\n", *sun4c_memerr_reg);
80         printk("REGISTER DUMP:\n");
81         show_regs(regs);
82         prom_halt();
83 }
84
85 static void unhandled_fault(unsigned long, struct task_struct *,
86                 struct pt_regs *) __attribute__ ((noreturn));
87
88 static void unhandled_fault(unsigned long address, struct task_struct *tsk,
89                      struct pt_regs *regs)
90 {
91         if((unsigned long) address < PAGE_SIZE) {
92                 printk(KERN_ALERT
93                     "Unable to handle kernel NULL pointer dereference\n");
94         } else {
95                 printk(KERN_ALERT "Unable to handle kernel paging request "
96                        "at virtual address %08lx\n", address);
97         }
98         printk(KERN_ALERT "tsk->{mm,active_mm}->context = %08lx\n",
99                 (tsk->mm ? tsk->mm->context : tsk->active_mm->context));
100         printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %08lx\n",
101                 (tsk->mm ? (unsigned long) tsk->mm->pgd :
102                         (unsigned long) tsk->active_mm->pgd));
103         die_if_kernel("Oops", regs);
104 }
105
106 asmlinkage int lookup_fault(unsigned long pc, unsigned long ret_pc, 
107                             unsigned long address)
108 {
109         struct pt_regs regs;
110         unsigned long g2;
111         unsigned int insn;
112         int i;
113         
114         i = search_extables_range(ret_pc, &g2);
115         switch (i) {
116         case 3:
117                 /* load & store will be handled by fixup */
118                 return 3;
119
120         case 1:
121                 /* store will be handled by fixup, load will bump out */
122                 /* for _to_ macros */
123                 insn = *((unsigned int *) pc);
124                 if ((insn >> 21) & 1)
125                         return 1;
126                 break;
127
128         case 2:
129                 /* load will be handled by fixup, store will bump out */
130                 /* for _from_ macros */
131                 insn = *((unsigned int *) pc);
132                 if (!((insn >> 21) & 1) || ((insn>>19)&0x3f) == 15)
133                         return 2; 
134                 break; 
135
136         default:
137                 break;
138         }
139
140         memset(&regs, 0, sizeof (regs));
141         regs.pc = pc;
142         regs.npc = pc + 4;
143         __asm__ __volatile__(
144                 "rd %%psr, %0\n\t"
145                 "nop\n\t"
146                 "nop\n\t"
147                 "nop\n" : "=r" (regs.psr));
148         unhandled_fault(address, current, &regs);
149
150         /* Not reached */
151         return 0;
152 }
153
154 static inline void
155 show_signal_msg(struct pt_regs *regs, int sig, int code,
156                 unsigned long address, struct task_struct *tsk)
157 {
158         if (!unhandled_signal(tsk, sig))
159                 return;
160
161         if (!printk_ratelimit())
162                 return;
163
164         printk("%s%s[%d]: segfault at %lx ip %p (rpc %p) sp %p error %x",
165                task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
166                tsk->comm, task_pid_nr(tsk), address,
167                (void *)regs->pc, (void *)regs->u_regs[UREG_I7],
168                (void *)regs->u_regs[UREG_FP], code);
169
170         print_vma_addr(KERN_CONT " in ", regs->pc);
171
172         printk(KERN_CONT "\n");
173 }
174
175 static void __do_fault_siginfo(int code, int sig, struct pt_regs *regs,
176                                unsigned long addr)
177 {
178         siginfo_t info;
179
180         info.si_signo = sig;
181         info.si_code = code;
182         info.si_errno = 0;
183         info.si_addr = (void __user *) addr;
184         info.si_trapno = 0;
185
186         if (unlikely(show_unhandled_signals))
187                 show_signal_msg(regs, sig, info.si_code,
188                                 addr, current);
189
190         force_sig_info (sig, &info, current);
191 }
192
193 extern unsigned long safe_compute_effective_address(struct pt_regs *,
194                                                     unsigned int);
195
196 static unsigned long compute_si_addr(struct pt_regs *regs, int text_fault)
197 {
198         unsigned int insn;
199
200         if (text_fault)
201                 return regs->pc;
202
203         if (regs->psr & PSR_PS) {
204                 insn = *(unsigned int *) regs->pc;
205         } else {
206                 __get_user(insn, (unsigned int *) regs->pc);
207         }
208
209         return safe_compute_effective_address(regs, insn);
210 }
211
212 static noinline void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
213                                       int text_fault)
214 {
215         unsigned long addr = compute_si_addr(regs, text_fault);
216
217         __do_fault_siginfo(code, sig, regs, addr);
218 }
219
220 asmlinkage void do_sparc_fault(struct pt_regs *regs, int text_fault, int write,
221                                unsigned long address)
222 {
223         struct vm_area_struct *vma;
224         struct task_struct *tsk = current;
225         struct mm_struct *mm = tsk->mm;
226         unsigned int fixup;
227         unsigned long g2;
228         int from_user = !(regs->psr & PSR_PS);
229         int fault, code;
230
231         if(text_fault)
232                 address = regs->pc;
233
234         /*
235          * We fault-in kernel-space virtual memory on-demand. The
236          * 'reference' page table is init_mm.pgd.
237          *
238          * NOTE! We MUST NOT take any locks for this case. We may
239          * be in an interrupt or a critical region, and should
240          * only copy the information from the master page table,
241          * nothing more.
242          */
243         code = SEGV_MAPERR;
244         if (!ARCH_SUN4C && address >= TASK_SIZE)
245                 goto vmalloc_fault;
246
247         /*
248          * If we're in an interrupt or have no user
249          * context, we must not take the fault..
250          */
251         if (in_atomic() || !mm)
252                 goto no_context;
253
254         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
255
256         down_read(&mm->mmap_sem);
257
258         /*
259          * The kernel referencing a bad kernel pointer can lock up
260          * a sun4c machine completely, so we must attempt recovery.
261          */
262         if(!from_user && address >= PAGE_OFFSET)
263                 goto bad_area;
264
265         vma = find_vma(mm, address);
266         if(!vma)
267                 goto bad_area;
268         if(vma->vm_start <= address)
269                 goto good_area;
270         if(!(vma->vm_flags & VM_GROWSDOWN))
271                 goto bad_area;
272         if(expand_stack(vma, address))
273                 goto bad_area;
274         /*
275          * Ok, we have a good vm_area for this memory access, so
276          * we can handle it..
277          */
278 good_area:
279         code = SEGV_ACCERR;
280         if(write) {
281                 if(!(vma->vm_flags & VM_WRITE))
282                         goto bad_area;
283         } else {
284                 /* Allow reads even for write-only mappings */
285                 if(!(vma->vm_flags & (VM_READ | VM_EXEC)))
286                         goto bad_area;
287         }
288
289         /*
290          * If for any reason at all we couldn't handle the fault,
291          * make sure we exit gracefully rather than endlessly redo
292          * the fault.
293          */
294         fault = handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0);
295         if (unlikely(fault & VM_FAULT_ERROR)) {
296                 if (fault & VM_FAULT_OOM)
297                         goto out_of_memory;
298                 else if (fault & VM_FAULT_SIGBUS)
299                         goto do_sigbus;
300                 BUG();
301         }
302         if (fault & VM_FAULT_MAJOR) {
303                 current->maj_flt++;
304                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
305         } else {
306                 current->min_flt++;
307                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
308         }
309         up_read(&mm->mmap_sem);
310         return;
311
312         /*
313          * Something tried to access memory that isn't in our memory map..
314          * Fix it, but check if it's kernel or user first..
315          */
316 bad_area:
317         up_read(&mm->mmap_sem);
318
319 bad_area_nosemaphore:
320         /* User mode accesses just cause a SIGSEGV */
321         if (from_user) {
322                 do_fault_siginfo(code, SIGSEGV, regs, text_fault);
323                 return;
324         }
325
326         /* Is this in ex_table? */
327 no_context:
328         g2 = regs->u_regs[UREG_G2];
329         if (!from_user) {
330                 fixup = search_extables_range(regs->pc, &g2);
331                 if (fixup > 10) { /* Values below are reserved for other things */
332                         extern const unsigned __memset_start[];
333                         extern const unsigned __memset_end[];
334                         extern const unsigned __csum_partial_copy_start[];
335                         extern const unsigned __csum_partial_copy_end[];
336
337 #ifdef DEBUG_EXCEPTIONS
338                         printk("Exception: PC<%08lx> faddr<%08lx>\n", regs->pc, address);
339                         printk("EX_TABLE: insn<%08lx> fixup<%08x> g2<%08lx>\n",
340                                 regs->pc, fixup, g2);
341 #endif
342                         if ((regs->pc >= (unsigned long)__memset_start &&
343                              regs->pc < (unsigned long)__memset_end) ||
344                             (regs->pc >= (unsigned long)__csum_partial_copy_start &&
345                              regs->pc < (unsigned long)__csum_partial_copy_end)) {
346                                 regs->u_regs[UREG_I4] = address;
347                                 regs->u_regs[UREG_I5] = regs->pc;
348                         }
349                         regs->u_regs[UREG_G2] = g2;
350                         regs->pc = fixup;
351                         regs->npc = regs->pc + 4;
352                         return;
353                 }
354         }
355         
356         unhandled_fault (address, tsk, regs);
357         do_exit(SIGKILL);
358
359 /*
360  * We ran out of memory, or some other thing happened to us that made
361  * us unable to handle the page fault gracefully.
362  */
363 out_of_memory:
364         up_read(&mm->mmap_sem);
365         if (from_user) {
366                 pagefault_out_of_memory();
367                 return;
368         }
369         goto no_context;
370
371 do_sigbus:
372         up_read(&mm->mmap_sem);
373         do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, text_fault);
374         if (!from_user)
375                 goto no_context;
376
377 vmalloc_fault:
378         {
379                 /*
380                  * Synchronize this task's top level page-table
381                  * with the 'reference' page table.
382                  */
383                 int offset = pgd_index(address);
384                 pgd_t *pgd, *pgd_k;
385                 pmd_t *pmd, *pmd_k;
386
387                 pgd = tsk->active_mm->pgd + offset;
388                 pgd_k = init_mm.pgd + offset;
389
390                 if (!pgd_present(*pgd)) {
391                         if (!pgd_present(*pgd_k))
392                                 goto bad_area_nosemaphore;
393                         pgd_val(*pgd) = pgd_val(*pgd_k);
394                         return;
395                 }
396
397                 pmd = pmd_offset(pgd, address);
398                 pmd_k = pmd_offset(pgd_k, address);
399
400                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
401                         goto bad_area_nosemaphore;
402                 *pmd = *pmd_k;
403                 return;
404         }
405 }
406
407 asmlinkage void do_sun4c_fault(struct pt_regs *regs, int text_fault, int write,
408                                unsigned long address)
409 {
410         extern void sun4c_update_mmu_cache(struct vm_area_struct *,
411                                            unsigned long,pte_t *);
412         extern pte_t *sun4c_pte_offset_kernel(pmd_t *,unsigned long);
413         struct task_struct *tsk = current;
414         struct mm_struct *mm = tsk->mm;
415         pgd_t *pgdp;
416         pte_t *ptep;
417
418         if (text_fault) {
419                 address = regs->pc;
420         } else if (!write &&
421                    !(regs->psr & PSR_PS)) {
422                 unsigned int insn, __user *ip;
423
424                 ip = (unsigned int __user *)regs->pc;
425                 if (!get_user(insn, ip)) {
426                         if ((insn & 0xc1680000) == 0xc0680000)
427                                 write = 1;
428                 }
429         }
430
431         if (!mm) {
432                 /* We are oopsing. */
433                 do_sparc_fault(regs, text_fault, write, address);
434                 BUG();  /* P3 Oops already, you bitch */
435         }
436
437         pgdp = pgd_offset(mm, address);
438         ptep = sun4c_pte_offset_kernel((pmd_t *) pgdp, address);
439
440         if (pgd_val(*pgdp)) {
441             if (write) {
442                 if ((pte_val(*ptep) & (_SUN4C_PAGE_WRITE|_SUN4C_PAGE_PRESENT))
443                                    == (_SUN4C_PAGE_WRITE|_SUN4C_PAGE_PRESENT)) {
444                         unsigned long flags;
445
446                         *ptep = __pte(pte_val(*ptep) | _SUN4C_PAGE_ACCESSED |
447                                       _SUN4C_PAGE_MODIFIED |
448                                       _SUN4C_PAGE_VALID |
449                                       _SUN4C_PAGE_DIRTY);
450
451                         local_irq_save(flags);
452                         if (sun4c_get_segmap(address) != invalid_segment) {
453                                 sun4c_put_pte(address, pte_val(*ptep));
454                                 local_irq_restore(flags);
455                                 return;
456                         }
457                         local_irq_restore(flags);
458                 }
459             } else {
460                 if ((pte_val(*ptep) & (_SUN4C_PAGE_READ|_SUN4C_PAGE_PRESENT))
461                                    == (_SUN4C_PAGE_READ|_SUN4C_PAGE_PRESENT)) {
462                         unsigned long flags;
463
464                         *ptep = __pte(pte_val(*ptep) | _SUN4C_PAGE_ACCESSED |
465                                       _SUN4C_PAGE_VALID);
466
467                         local_irq_save(flags);
468                         if (sun4c_get_segmap(address) != invalid_segment) {
469                                 sun4c_put_pte(address, pte_val(*ptep));
470                                 local_irq_restore(flags);
471                                 return;
472                         }
473                         local_irq_restore(flags);
474                 }
475             }
476         }
477
478         /* This conditional is 'interesting'. */
479         if (pgd_val(*pgdp) && !(write && !(pte_val(*ptep) & _SUN4C_PAGE_WRITE))
480             && (pte_val(*ptep) & _SUN4C_PAGE_VALID))
481                 /* Note: It is safe to not grab the MMAP semaphore here because
482                  *       we know that update_mmu_cache() will not sleep for
483                  *       any reason (at least not in the current implementation)
484                  *       and therefore there is no danger of another thread getting
485                  *       on the CPU and doing a shrink_mmap() on this vma.
486                  */
487                 sun4c_update_mmu_cache (find_vma(current->mm, address), address,
488                                         ptep);
489         else
490                 do_sparc_fault(regs, text_fault, write, address);
491 }
492
493 /* This always deals with user addresses. */
494 static void force_user_fault(unsigned long address, int write)
495 {
496         struct vm_area_struct *vma;
497         struct task_struct *tsk = current;
498         struct mm_struct *mm = tsk->mm;
499         int code;
500
501         code = SEGV_MAPERR;
502
503         down_read(&mm->mmap_sem);
504         vma = find_vma(mm, address);
505         if(!vma)
506                 goto bad_area;
507         if(vma->vm_start <= address)
508                 goto good_area;
509         if(!(vma->vm_flags & VM_GROWSDOWN))
510                 goto bad_area;
511         if(expand_stack(vma, address))
512                 goto bad_area;
513 good_area:
514         code = SEGV_ACCERR;
515         if(write) {
516                 if(!(vma->vm_flags & VM_WRITE))
517                         goto bad_area;
518         } else {
519                 if(!(vma->vm_flags & (VM_READ | VM_EXEC)))
520                         goto bad_area;
521         }
522         switch (handle_mm_fault(mm, vma, address, write ? FAULT_FLAG_WRITE : 0)) {
523         case VM_FAULT_SIGBUS:
524         case VM_FAULT_OOM:
525                 goto do_sigbus;
526         }
527         up_read(&mm->mmap_sem);
528         return;
529 bad_area:
530         up_read(&mm->mmap_sem);
531         __do_fault_siginfo(code, SIGSEGV, tsk->thread.kregs, address);
532         return;
533
534 do_sigbus:
535         up_read(&mm->mmap_sem);
536         __do_fault_siginfo(BUS_ADRERR, SIGBUS, tsk->thread.kregs, address);
537 }
538
539 static void check_stack_aligned(unsigned long sp)
540 {
541         if (sp & 0x7UL)
542                 force_sig(SIGILL, current);
543 }
544
545 void window_overflow_fault(void)
546 {
547         unsigned long sp;
548
549         sp = current_thread_info()->rwbuf_stkptrs[0];
550         if(((sp + 0x38) & PAGE_MASK) != (sp & PAGE_MASK))
551                 force_user_fault(sp + 0x38, 1);
552         force_user_fault(sp, 1);
553
554         check_stack_aligned(sp);
555 }
556
557 void window_underflow_fault(unsigned long sp)
558 {
559         if(((sp + 0x38) & PAGE_MASK) != (sp & PAGE_MASK))
560                 force_user_fault(sp + 0x38, 0);
561         force_user_fault(sp, 0);
562
563         check_stack_aligned(sp);
564 }
565
566 void window_ret_fault(struct pt_regs *regs)
567 {
568         unsigned long sp;
569
570         sp = regs->u_regs[UREG_FP];
571         if(((sp + 0x38) & PAGE_MASK) != (sp & PAGE_MASK))
572                 force_user_fault(sp + 0x38, 0);
573         force_user_fault(sp, 0);
574
575         check_stack_aligned(sp);
576 }