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/vt_kern.h> /* For unblank_screen() */ 20 #include <linux/module.h> 21 22 #include <asm/branch.h> 23 #include <asm/mmu_context.h> 24 #include <asm/system.h> 25 #include <asm/uaccess.h> 26 #include <asm/ptrace.h> 27 #include <asm/highmem.h> /* For VMALLOC_END */ 28 29 /* 30 * This routine handles page faults. It determines the address, 31 * and the problem, and then passes it off to one of the appropriate 32 * routines. 33 */ 34 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long write, 35 unsigned long address) 36 { 37 struct vm_area_struct * vma = NULL; 38 struct task_struct *tsk = current; 39 struct mm_struct *mm = tsk->mm; 40 const int field = sizeof(unsigned long) * 2; 41 siginfo_t info; 42 int fault; 43 44 #if 0 45 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_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 */ 61 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END)) 62 goto vmalloc_fault; 63 #ifdef MODULE_START 64 if (unlikely(address >= MODULE_START && address < MODULE_END)) 65 goto vmalloc_fault; 66 #endif 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 { 96 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))) 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 fault = handle_mm_fault(mm, vma, address, write); 107 if (unlikely(fault & VM_FAULT_ERROR)) { 108 if (fault & VM_FAULT_OOM) 109 goto out_of_memory; 110 else if (fault & VM_FAULT_SIGBUS) 111 goto do_sigbus; 112 BUG(); 113 } 114 if (fault & VM_FAULT_MAJOR) 115 tsk->maj_flt++; 116 else 117 tsk->min_flt++; 118 119 up_read(&mm->mmap_sem); 120 return; 121 122 /* 123 * Something tried to access memory that isn't in our memory map.. 124 * Fix it, but check if it's kernel or user first.. 125 */ 126 bad_area: 127 up_read(&mm->mmap_sem); 128 129 bad_area_nosemaphore: 130 /* User mode accesses just cause a SIGSEGV */ 131 if (user_mode(regs)) { 132 tsk->thread.cp0_badvaddr = address; 133 tsk->thread.error_code = write; 134 #if 0 135 printk("do_page_fault() #2: sending SIGSEGV to %s for " 136 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n", 137 tsk->comm, 138 write ? "write access to" : "read access from", 139 field, address, 140 field, (unsigned long) regs->cp0_epc, 141 field, (unsigned long) regs->regs[31]); 142 #endif 143 info.si_signo = SIGSEGV; 144 info.si_errno = 0; 145 /* info.si_code has been set above */ 146 info.si_addr = (void __user *) address; 147 force_sig_info(SIGSEGV, &info, tsk); 148 return; 149 } 150 151 no_context: 152 /* Are we prepared to handle this kernel fault? */ 153 if (fixup_exception(regs)) { 154 current->thread.cp0_baduaddr = address; 155 return; 156 } 157 158 /* 159 * Oops. The kernel tried to access some bad page. We'll have to 160 * terminate things with extreme prejudice. 161 */ 162 bust_spinlocks(1); 163 164 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at " 165 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n", 166 raw_smp_processor_id(), field, address, field, regs->cp0_epc, 167 field, regs->regs[31]); 168 die("Oops", regs); 169 170 /* 171 * We ran out of memory, or some other thing happened to us that made 172 * us unable to handle the page fault gracefully. 173 */ 174 out_of_memory: 175 up_read(&mm->mmap_sem); 176 if (is_global_init(tsk)) { 177 yield(); 178 down_read(&mm->mmap_sem); 179 goto survive; 180 } 181 printk("VM: killing process %s\n", tsk->comm); 182 if (user_mode(regs)) 183 do_group_exit(SIGKILL); 184 goto no_context; 185 186 do_sigbus: 187 up_read(&mm->mmap_sem); 188 189 /* Kernel mode? Handle exceptions or die */ 190 if (!user_mode(regs)) 191 goto no_context; 192 else 193 /* 194 * Send a sigbus, regardless of whether we were in kernel 195 * or user mode. 196 */ 197 #if 0 198 printk("do_page_fault() #3: sending SIGBUS to %s for " 199 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n", 200 tsk->comm, 201 write ? "write access to" : "read access from", 202 field, address, 203 field, (unsigned long) regs->cp0_epc, 204 field, (unsigned long) regs->regs[31]); 205 #endif 206 tsk->thread.cp0_badvaddr = address; 207 info.si_signo = SIGBUS; 208 info.si_errno = 0; 209 info.si_code = BUS_ADRERR; 210 info.si_addr = (void __user *) address; 211 force_sig_info(SIGBUS, &info, tsk); 212 213 return; 214 vmalloc_fault: 215 { 216 /* 217 * Synchronize this task's top level page-table 218 * with the 'reference' page table. 219 * 220 * Do _not_ use "tsk" here. We might be inside 221 * an interrupt in the middle of a task switch.. 222 */ 223 int offset = __pgd_offset(address); 224 pgd_t *pgd, *pgd_k; 225 pud_t *pud, *pud_k; 226 pmd_t *pmd, *pmd_k; 227 pte_t *pte_k; 228 229 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset; 230 pgd_k = init_mm.pgd + offset; 231 232 if (!pgd_present(*pgd_k)) 233 goto no_context; 234 set_pgd(pgd, *pgd_k); 235 236 pud = pud_offset(pgd, address); 237 pud_k = pud_offset(pgd_k, address); 238 if (!pud_present(*pud_k)) 239 goto no_context; 240 241 pmd = pmd_offset(pud, address); 242 pmd_k = pmd_offset(pud_k, address); 243 if (!pmd_present(*pmd_k)) 244 goto no_context; 245 set_pmd(pmd, *pmd_k); 246 247 pte_k = pte_offset_kernel(pmd_k, address); 248 if (!pte_present(*pte_k)) 249 goto no_context; 250 return; 251 } 252 } 253