Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/arch/alpha/mm/fault.c |
| 3 | * |
| 4 | * Copyright (C) 1995 Linus Torvalds |
| 5 | */ |
| 6 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | #include <linux/sched.h> |
| 8 | #include <linux/kernel.h> |
| 9 | #include <linux/mm.h> |
| 10 | #include <asm/io.h> |
| 11 | |
| 12 | #define __EXTERN_INLINE inline |
| 13 | #include <asm/mmu_context.h> |
| 14 | #include <asm/tlbflush.h> |
| 15 | #undef __EXTERN_INLINE |
| 16 | |
| 17 | #include <linux/signal.h> |
| 18 | #include <linux/errno.h> |
| 19 | #include <linux/string.h> |
| 20 | #include <linux/types.h> |
| 21 | #include <linux/ptrace.h> |
| 22 | #include <linux/mman.h> |
| 23 | #include <linux/smp.h> |
| 24 | #include <linux/smp_lock.h> |
| 25 | #include <linux/interrupt.h> |
| 26 | #include <linux/module.h> |
| 27 | |
| 28 | #include <asm/system.h> |
| 29 | #include <asm/uaccess.h> |
| 30 | |
| 31 | extern void die_if_kernel(char *,struct pt_regs *,long, unsigned long *); |
| 32 | |
| 33 | |
| 34 | /* |
| 35 | * Force a new ASN for a task. |
| 36 | */ |
| 37 | |
| 38 | #ifndef CONFIG_SMP |
| 39 | unsigned long last_asn = ASN_FIRST_VERSION; |
| 40 | #endif |
| 41 | |
| 42 | void |
| 43 | __load_new_mm_context(struct mm_struct *next_mm) |
| 44 | { |
| 45 | unsigned long mmc; |
| 46 | struct pcb_struct *pcb; |
| 47 | |
| 48 | mmc = __get_new_mm_context(next_mm, smp_processor_id()); |
| 49 | next_mm->context[smp_processor_id()] = mmc; |
| 50 | |
| 51 | pcb = ¤t_thread_info()->pcb; |
| 52 | pcb->asn = mmc & HARDWARE_ASN_MASK; |
| 53 | pcb->ptbr = ((unsigned long) next_mm->pgd - IDENT_ADDR) >> PAGE_SHIFT; |
| 54 | |
| 55 | __reload_thread(pcb); |
| 56 | } |
| 57 | |
| 58 | |
| 59 | /* |
| 60 | * This routine handles page faults. It determines the address, |
| 61 | * and the problem, and then passes it off to handle_mm_fault(). |
| 62 | * |
| 63 | * mmcsr: |
| 64 | * 0 = translation not valid |
| 65 | * 1 = access violation |
| 66 | * 2 = fault-on-read |
| 67 | * 3 = fault-on-execute |
| 68 | * 4 = fault-on-write |
| 69 | * |
| 70 | * cause: |
| 71 | * -1 = instruction fetch |
| 72 | * 0 = load |
| 73 | * 1 = store |
| 74 | * |
| 75 | * Registers $9 through $15 are saved in a block just prior to `regs' and |
| 76 | * are saved and restored around the call to allow exception code to |
| 77 | * modify them. |
| 78 | */ |
| 79 | |
| 80 | /* Macro for exception fixup code to access integer registers. */ |
| 81 | #define dpf_reg(r) \ |
| 82 | (((unsigned long *)regs)[(r) <= 8 ? (r) : (r) <= 15 ? (r)-16 : \ |
| 83 | (r) <= 18 ? (r)+8 : (r)-10]) |
| 84 | |
| 85 | asmlinkage void |
| 86 | do_page_fault(unsigned long address, unsigned long mmcsr, |
| 87 | long cause, struct pt_regs *regs) |
| 88 | { |
| 89 | struct vm_area_struct * vma; |
| 90 | struct mm_struct *mm = current->mm; |
| 91 | const struct exception_table_entry *fixup; |
| 92 | int fault, si_code = SEGV_MAPERR; |
| 93 | siginfo_t info; |
| 94 | |
| 95 | /* As of EV6, a load into $31/$f31 is a prefetch, and never faults |
| 96 | (or is suppressed by the PALcode). Support that for older CPUs |
| 97 | by ignoring such an instruction. */ |
| 98 | if (cause == 0) { |
| 99 | unsigned int insn; |
| 100 | __get_user(insn, (unsigned int __user *)regs->pc); |
| 101 | if ((insn >> 21 & 0x1f) == 0x1f && |
| 102 | /* ldq ldl ldt lds ldg ldf ldwu ldbu */ |
| 103 | (1ul << (insn >> 26) & 0x30f00001400ul)) { |
| 104 | regs->pc += 4; |
| 105 | return; |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | /* If we're in an interrupt context, or have no user context, |
| 110 | we must not take the fault. */ |
Peter Zijlstra | 6edaf68 | 2006-12-06 20:32:18 -0800 | [diff] [blame^] | 111 | if (!mm || in_atomic()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | goto no_context; |
| 113 | |
| 114 | #ifdef CONFIG_ALPHA_LARGE_VMALLOC |
| 115 | if (address >= TASK_SIZE) |
| 116 | goto vmalloc_fault; |
| 117 | #endif |
| 118 | |
| 119 | down_read(&mm->mmap_sem); |
| 120 | vma = find_vma(mm, address); |
| 121 | if (!vma) |
| 122 | goto bad_area; |
| 123 | if (vma->vm_start <= address) |
| 124 | goto good_area; |
| 125 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
| 126 | goto bad_area; |
| 127 | if (expand_stack(vma, address)) |
| 128 | goto bad_area; |
| 129 | |
| 130 | /* Ok, we have a good vm_area for this memory access, so |
| 131 | we can handle it. */ |
| 132 | good_area: |
| 133 | si_code = SEGV_ACCERR; |
| 134 | if (cause < 0) { |
| 135 | if (!(vma->vm_flags & VM_EXEC)) |
| 136 | goto bad_area; |
| 137 | } else if (!cause) { |
| 138 | /* Allow reads even for write-only mappings */ |
| 139 | if (!(vma->vm_flags & (VM_READ | VM_WRITE))) |
| 140 | goto bad_area; |
| 141 | } else { |
| 142 | if (!(vma->vm_flags & VM_WRITE)) |
| 143 | goto bad_area; |
| 144 | } |
| 145 | |
| 146 | survive: |
| 147 | /* If for any reason at all we couldn't handle the fault, |
| 148 | make sure we exit gracefully rather than endlessly redo |
| 149 | the fault. */ |
| 150 | fault = handle_mm_fault(mm, vma, address, cause > 0); |
| 151 | up_read(&mm->mmap_sem); |
| 152 | |
| 153 | switch (fault) { |
| 154 | case VM_FAULT_MINOR: |
| 155 | current->min_flt++; |
| 156 | break; |
| 157 | case VM_FAULT_MAJOR: |
| 158 | current->maj_flt++; |
| 159 | break; |
| 160 | case VM_FAULT_SIGBUS: |
| 161 | goto do_sigbus; |
| 162 | case VM_FAULT_OOM: |
| 163 | goto out_of_memory; |
| 164 | default: |
| 165 | BUG(); |
| 166 | } |
| 167 | return; |
| 168 | |
| 169 | /* Something tried to access memory that isn't in our memory map. |
| 170 | Fix it, but check if it's kernel or user first. */ |
| 171 | bad_area: |
| 172 | up_read(&mm->mmap_sem); |
| 173 | |
| 174 | if (user_mode(regs)) |
| 175 | goto do_sigsegv; |
| 176 | |
| 177 | no_context: |
| 178 | /* Are we prepared to handle this fault as an exception? */ |
| 179 | if ((fixup = search_exception_tables(regs->pc)) != 0) { |
| 180 | unsigned long newpc; |
| 181 | newpc = fixup_exception(dpf_reg, fixup, regs->pc); |
| 182 | regs->pc = newpc; |
| 183 | return; |
| 184 | } |
| 185 | |
| 186 | /* Oops. The kernel tried to access some bad page. We'll have to |
| 187 | terminate things with extreme prejudice. */ |
| 188 | printk(KERN_ALERT "Unable to handle kernel paging request at " |
| 189 | "virtual address %016lx\n", address); |
| 190 | die_if_kernel("Oops", regs, cause, (unsigned long*)regs - 16); |
| 191 | do_exit(SIGKILL); |
| 192 | |
| 193 | /* We ran out of memory, or some other thing happened to us that |
| 194 | made us unable to handle the page fault gracefully. */ |
| 195 | out_of_memory: |
Sukadev Bhattiprolu | f400e19 | 2006-09-29 02:00:07 -0700 | [diff] [blame] | 196 | if (is_init(current)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | yield(); |
| 198 | down_read(&mm->mmap_sem); |
| 199 | goto survive; |
| 200 | } |
| 201 | printk(KERN_ALERT "VM: killing process %s(%d)\n", |
| 202 | current->comm, current->pid); |
| 203 | if (!user_mode(regs)) |
| 204 | goto no_context; |
| 205 | do_exit(SIGKILL); |
| 206 | |
| 207 | do_sigbus: |
| 208 | /* Send a sigbus, regardless of whether we were in kernel |
| 209 | or user mode. */ |
| 210 | info.si_signo = SIGBUS; |
| 211 | info.si_errno = 0; |
| 212 | info.si_code = BUS_ADRERR; |
| 213 | info.si_addr = (void __user *) address; |
| 214 | force_sig_info(SIGBUS, &info, current); |
| 215 | if (!user_mode(regs)) |
| 216 | goto no_context; |
| 217 | return; |
| 218 | |
| 219 | do_sigsegv: |
| 220 | info.si_signo = SIGSEGV; |
| 221 | info.si_errno = 0; |
| 222 | info.si_code = si_code; |
| 223 | info.si_addr = (void __user *) address; |
| 224 | force_sig_info(SIGSEGV, &info, current); |
| 225 | return; |
| 226 | |
| 227 | #ifdef CONFIG_ALPHA_LARGE_VMALLOC |
| 228 | vmalloc_fault: |
| 229 | if (user_mode(regs)) |
| 230 | goto do_sigsegv; |
| 231 | else { |
| 232 | /* Synchronize this task's top level page-table |
| 233 | with the "reference" page table from init. */ |
| 234 | long index = pgd_index(address); |
| 235 | pgd_t *pgd, *pgd_k; |
| 236 | |
| 237 | pgd = current->active_mm->pgd + index; |
| 238 | pgd_k = swapper_pg_dir + index; |
| 239 | if (!pgd_present(*pgd) && pgd_present(*pgd_k)) { |
| 240 | pgd_val(*pgd) = pgd_val(*pgd_k); |
| 241 | return; |
| 242 | } |
| 243 | goto no_context; |
| 244 | } |
| 245 | #endif |
| 246 | } |