Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * This file is subject to the terms and conditions of the GNU General Public |
| 3 | * License. See the file "COPYING" in the main directory of this archive |
| 4 | * for more details. |
| 5 | * |
| 6 | * Copyright (C) 1995 - 2000 by Ralf Baechle |
| 7 | */ |
| 8 | #include <linux/signal.h> |
| 9 | #include <linux/sched.h> |
| 10 | #include <linux/interrupt.h> |
| 11 | #include <linux/kernel.h> |
| 12 | #include <linux/errno.h> |
| 13 | #include <linux/string.h> |
| 14 | #include <linux/types.h> |
| 15 | #include <linux/ptrace.h> |
| 16 | #include <linux/mman.h> |
| 17 | #include <linux/mm.h> |
| 18 | #include <linux/smp.h> |
| 19 | #include <linux/smp_lock.h> |
| 20 | #include <linux/vt_kern.h> /* For unblank_screen() */ |
| 21 | #include <linux/module.h> |
| 22 | |
| 23 | #include <asm/branch.h> |
| 24 | #include <asm/mmu_context.h> |
| 25 | #include <asm/system.h> |
| 26 | #include <asm/uaccess.h> |
| 27 | #include <asm/ptrace.h> |
Thiemo Seufer | 16033d6 | 2005-02-19 13:56:04 +0000 | [diff] [blame] | 28 | #include <asm/highmem.h> /* For VMALLOC_END */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | |
| 30 | /* |
| 31 | * This routine handles page faults. It determines the address, |
| 32 | * and the problem, and then passes it off to one of the appropriate |
| 33 | * routines. |
| 34 | */ |
| 35 | asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long write, |
| 36 | unsigned long address) |
| 37 | { |
| 38 | struct vm_area_struct * vma = NULL; |
| 39 | struct task_struct *tsk = current; |
| 40 | struct mm_struct *mm = tsk->mm; |
| 41 | const int field = sizeof(unsigned long) * 2; |
| 42 | siginfo_t info; |
| 43 | |
| 44 | #if 0 |
| 45 | printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", smp_processor_id(), |
| 46 | current->comm, current->pid, field, address, write, |
| 47 | field, regs->cp0_epc); |
| 48 | #endif |
| 49 | |
| 50 | info.si_code = SEGV_MAPERR; |
| 51 | |
| 52 | /* |
| 53 | * We fault-in kernel-space virtual memory on-demand. The |
| 54 | * 'reference' page table is init_mm.pgd. |
| 55 | * |
| 56 | * NOTE! We MUST NOT take any locks for this case. We may |
| 57 | * be in an interrupt or a critical region, and should |
| 58 | * only copy the information from the master page table, |
| 59 | * nothing more. |
| 60 | */ |
Thiemo Seufer | 16033d6 | 2005-02-19 13:56:04 +0000 | [diff] [blame] | 61 | if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | goto vmalloc_fault; |
Atsushi Nemoto | 656be92 | 2006-10-26 00:08:31 +0900 | [diff] [blame^] | 63 | #ifdef MODULE_START |
| 64 | if (unlikely(address >= MODULE_START && address < MODULE_END)) |
| 65 | goto vmalloc_fault; |
| 66 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | |
| 68 | /* |
| 69 | * If we're in an interrupt or have no user |
| 70 | * context, we must not take the fault.. |
| 71 | */ |
| 72 | if (in_atomic() || !mm) |
| 73 | goto bad_area_nosemaphore; |
| 74 | |
| 75 | down_read(&mm->mmap_sem); |
| 76 | vma = find_vma(mm, address); |
| 77 | if (!vma) |
| 78 | goto bad_area; |
| 79 | if (vma->vm_start <= address) |
| 80 | goto good_area; |
| 81 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
| 82 | goto bad_area; |
| 83 | if (expand_stack(vma, address)) |
| 84 | goto bad_area; |
| 85 | /* |
| 86 | * Ok, we have a good vm_area for this memory access, so |
| 87 | * we can handle it.. |
| 88 | */ |
| 89 | good_area: |
| 90 | info.si_code = SEGV_ACCERR; |
| 91 | |
| 92 | if (write) { |
| 93 | if (!(vma->vm_flags & VM_WRITE)) |
| 94 | goto bad_area; |
| 95 | } else { |
Ralf Baechle | 00932ba | 2006-09-16 01:29:37 +0100 | [diff] [blame] | 96 | if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | goto bad_area; |
| 98 | } |
| 99 | |
| 100 | survive: |
| 101 | /* |
| 102 | * If for any reason at all we couldn't handle the fault, |
| 103 | * make sure we exit gracefully rather than endlessly redo |
| 104 | * the fault. |
| 105 | */ |
| 106 | switch (handle_mm_fault(mm, vma, address, write)) { |
| 107 | case VM_FAULT_MINOR: |
| 108 | tsk->min_flt++; |
| 109 | break; |
| 110 | case VM_FAULT_MAJOR: |
| 111 | tsk->maj_flt++; |
| 112 | break; |
| 113 | case VM_FAULT_SIGBUS: |
| 114 | goto do_sigbus; |
| 115 | case VM_FAULT_OOM: |
| 116 | goto out_of_memory; |
| 117 | default: |
| 118 | BUG(); |
| 119 | } |
| 120 | |
| 121 | up_read(&mm->mmap_sem); |
| 122 | return; |
| 123 | |
| 124 | /* |
| 125 | * Something tried to access memory that isn't in our memory map.. |
| 126 | * Fix it, but check if it's kernel or user first.. |
| 127 | */ |
| 128 | bad_area: |
| 129 | up_read(&mm->mmap_sem); |
| 130 | |
| 131 | bad_area_nosemaphore: |
| 132 | /* User mode accesses just cause a SIGSEGV */ |
| 133 | if (user_mode(regs)) { |
| 134 | tsk->thread.cp0_badvaddr = address; |
| 135 | tsk->thread.error_code = write; |
| 136 | #if 0 |
| 137 | printk("do_page_fault() #2: sending SIGSEGV to %s for " |
| 138 | "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n", |
| 139 | tsk->comm, |
| 140 | write ? "write access to" : "read access from", |
| 141 | field, address, |
| 142 | field, (unsigned long) regs->cp0_epc, |
| 143 | field, (unsigned long) regs->regs[31]); |
| 144 | #endif |
| 145 | info.si_signo = SIGSEGV; |
| 146 | info.si_errno = 0; |
| 147 | /* info.si_code has been set above */ |
Ralf Baechle | fe00f94 | 2005-03-01 19:22:29 +0000 | [diff] [blame] | 148 | info.si_addr = (void __user *) address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | force_sig_info(SIGSEGV, &info, tsk); |
| 150 | return; |
| 151 | } |
| 152 | |
| 153 | no_context: |
| 154 | /* Are we prepared to handle this kernel fault? */ |
| 155 | if (fixup_exception(regs)) { |
| 156 | current->thread.cp0_baduaddr = address; |
| 157 | return; |
| 158 | } |
| 159 | |
| 160 | /* |
| 161 | * Oops. The kernel tried to access some bad page. We'll have to |
| 162 | * terminate things with extreme prejudice. |
| 163 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | bust_spinlocks(1); |
| 165 | |
| 166 | printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at " |
| 167 | "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n", |
| 168 | smp_processor_id(), field, address, field, regs->cp0_epc, |
| 169 | field, regs->regs[31]); |
| 170 | die("Oops", regs); |
| 171 | |
| 172 | /* |
| 173 | * We ran out of memory, or some other thing happened to us that made |
| 174 | * us unable to handle the page fault gracefully. |
| 175 | */ |
| 176 | out_of_memory: |
| 177 | up_read(&mm->mmap_sem); |
Sukadev Bhattiprolu | f400e19 | 2006-09-29 02:00:07 -0700 | [diff] [blame] | 178 | if (is_init(tsk)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | yield(); |
| 180 | down_read(&mm->mmap_sem); |
| 181 | goto survive; |
| 182 | } |
| 183 | printk("VM: killing process %s\n", tsk->comm); |
| 184 | if (user_mode(regs)) |
| 185 | do_exit(SIGKILL); |
| 186 | goto no_context; |
| 187 | |
| 188 | do_sigbus: |
| 189 | up_read(&mm->mmap_sem); |
| 190 | |
| 191 | /* Kernel mode? Handle exceptions or die */ |
| 192 | if (!user_mode(regs)) |
| 193 | goto no_context; |
Ralf Baechle | 41c594a | 2006-04-05 09:45:45 +0100 | [diff] [blame] | 194 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | /* |
| 196 | * Send a sigbus, regardless of whether we were in kernel |
| 197 | * or user mode. |
| 198 | */ |
Ralf Baechle | 41c594a | 2006-04-05 09:45:45 +0100 | [diff] [blame] | 199 | #if 0 |
| 200 | printk("do_page_fault() #3: sending SIGBUS to %s for " |
| 201 | "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n", |
| 202 | tsk->comm, |
| 203 | write ? "write access to" : "read access from", |
| 204 | field, address, |
| 205 | field, (unsigned long) regs->cp0_epc, |
| 206 | field, (unsigned long) regs->regs[31]); |
| 207 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | tsk->thread.cp0_badvaddr = address; |
| 209 | info.si_signo = SIGBUS; |
| 210 | info.si_errno = 0; |
| 211 | info.si_code = BUS_ADRERR; |
Ralf Baechle | fe00f94 | 2005-03-01 19:22:29 +0000 | [diff] [blame] | 212 | info.si_addr = (void __user *) address; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | force_sig_info(SIGBUS, &info, tsk); |
| 214 | |
| 215 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | vmalloc_fault: |
| 217 | { |
| 218 | /* |
| 219 | * Synchronize this task's top level page-table |
| 220 | * with the 'reference' page table. |
| 221 | * |
| 222 | * Do _not_ use "tsk" here. We might be inside |
| 223 | * an interrupt in the middle of a task switch.. |
| 224 | */ |
| 225 | int offset = __pgd_offset(address); |
| 226 | pgd_t *pgd, *pgd_k; |
Ralf Baechle | c6e8b58 | 2005-02-10 12:19:59 +0000 | [diff] [blame] | 227 | pud_t *pud, *pud_k; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | pmd_t *pmd, *pmd_k; |
| 229 | pte_t *pte_k; |
| 230 | |
| 231 | pgd = (pgd_t *) pgd_current[smp_processor_id()] + offset; |
| 232 | pgd_k = init_mm.pgd + offset; |
| 233 | |
| 234 | if (!pgd_present(*pgd_k)) |
| 235 | goto no_context; |
| 236 | set_pgd(pgd, *pgd_k); |
| 237 | |
Ralf Baechle | c6e8b58 | 2005-02-10 12:19:59 +0000 | [diff] [blame] | 238 | pud = pud_offset(pgd, address); |
| 239 | pud_k = pud_offset(pgd_k, address); |
| 240 | if (!pud_present(*pud_k)) |
| 241 | goto no_context; |
| 242 | |
| 243 | pmd = pmd_offset(pud, address); |
| 244 | pmd_k = pmd_offset(pud_k, address); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | if (!pmd_present(*pmd_k)) |
| 246 | goto no_context; |
| 247 | set_pmd(pmd, *pmd_k); |
| 248 | |
| 249 | pte_k = pte_offset_kernel(pmd_k, address); |
| 250 | if (!pte_present(*pte_k)) |
| 251 | goto no_context; |
| 252 | return; |
| 253 | } |
| 254 | } |