Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * MMU fault handling support. |
| 3 | * |
| 4 | * Copyright (C) 1998-2002 Hewlett-Packard Co |
| 5 | * David Mosberger-Tang <davidm@hpl.hp.com> |
| 6 | */ |
| 7 | #include <linux/sched.h> |
| 8 | #include <linux/kernel.h> |
| 9 | #include <linux/mm.h> |
| 10 | #include <linux/smp_lock.h> |
| 11 | #include <linux/interrupt.h> |
| 12 | |
| 13 | #include <asm/pgtable.h> |
| 14 | #include <asm/processor.h> |
| 15 | #include <asm/system.h> |
| 16 | #include <asm/uaccess.h> |
Anil S Keshavamurthy | 7213b25 | 2005-06-23 00:09:27 -0700 | [diff] [blame] | 17 | #include <asm/kdebug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | |
| 19 | extern void die (char *, struct pt_regs *, long); |
| 20 | |
| 21 | /* |
| 22 | * This routine is analogous to expand_stack() but instead grows the |
| 23 | * register backing store (which grows towards higher addresses). |
| 24 | * Since the register backing store is access sequentially, we |
| 25 | * disallow growing the RBS by more than a page at a time. Note that |
| 26 | * the VM_GROWSUP flag can be set on any VM area but that's fine |
| 27 | * because the total process size is still limited by RLIMIT_STACK and |
| 28 | * RLIMIT_AS. |
| 29 | */ |
| 30 | static inline long |
| 31 | expand_backing_store (struct vm_area_struct *vma, unsigned long address) |
| 32 | { |
| 33 | unsigned long grow; |
| 34 | |
| 35 | grow = PAGE_SIZE >> PAGE_SHIFT; |
| 36 | if (address - vma->vm_start > current->signal->rlim[RLIMIT_STACK].rlim_cur |
| 37 | || (((vma->vm_mm->total_vm + grow) << PAGE_SHIFT) > current->signal->rlim[RLIMIT_AS].rlim_cur)) |
| 38 | return -ENOMEM; |
| 39 | vma->vm_end += PAGE_SIZE; |
| 40 | vma->vm_mm->total_vm += grow; |
| 41 | if (vma->vm_flags & VM_LOCKED) |
| 42 | vma->vm_mm->locked_vm += grow; |
| 43 | __vm_stat_account(vma->vm_mm, vma->vm_flags, vma->vm_file, grow); |
| 44 | return 0; |
| 45 | } |
| 46 | |
| 47 | /* |
| 48 | * Return TRUE if ADDRESS points at a page in the kernel's mapped segment |
| 49 | * (inside region 5, on ia64) and that page is present. |
| 50 | */ |
| 51 | static int |
| 52 | mapped_kernel_page_is_present (unsigned long address) |
| 53 | { |
| 54 | pgd_t *pgd; |
| 55 | pud_t *pud; |
| 56 | pmd_t *pmd; |
| 57 | pte_t *ptep, pte; |
| 58 | |
| 59 | pgd = pgd_offset_k(address); |
| 60 | if (pgd_none(*pgd) || pgd_bad(*pgd)) |
| 61 | return 0; |
| 62 | |
| 63 | pud = pud_offset(pgd, address); |
| 64 | if (pud_none(*pud) || pud_bad(*pud)) |
| 65 | return 0; |
| 66 | |
| 67 | pmd = pmd_offset(pud, address); |
| 68 | if (pmd_none(*pmd) || pmd_bad(*pmd)) |
| 69 | return 0; |
| 70 | |
| 71 | ptep = pte_offset_kernel(pmd, address); |
| 72 | if (!ptep) |
| 73 | return 0; |
| 74 | |
| 75 | pte = *ptep; |
| 76 | return pte_present(pte); |
| 77 | } |
| 78 | |
| 79 | void |
| 80 | ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs) |
| 81 | { |
| 82 | int signal = SIGSEGV, code = SEGV_MAPERR; |
| 83 | struct vm_area_struct *vma, *prev_vma; |
| 84 | struct mm_struct *mm = current->mm; |
| 85 | struct siginfo si; |
| 86 | unsigned long mask; |
| 87 | |
| 88 | /* |
| 89 | * If we're in an interrupt or have no user context, we must not take the fault.. |
| 90 | */ |
| 91 | if (in_atomic() || !mm) |
| 92 | goto no_context; |
| 93 | |
| 94 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
| 95 | /* |
| 96 | * If fault is in region 5 and we are in the kernel, we may already |
| 97 | * have the mmap_sem (pfn_valid macro is called during mmap). There |
| 98 | * is no vma for region 5 addr's anyway, so skip getting the semaphore |
| 99 | * and go directly to the exception handling code. |
| 100 | */ |
| 101 | |
| 102 | if ((REGION_NUMBER(address) == 5) && !user_mode(regs)) |
| 103 | goto bad_area_no_up; |
| 104 | #endif |
| 105 | |
Anil S Keshavamurthy | 7213b25 | 2005-06-23 00:09:27 -0700 | [diff] [blame] | 106 | /* |
| 107 | * This is to handle the kprobes on user space access instructions |
| 108 | */ |
| 109 | if (notify_die(DIE_PAGE_FAULT, "page fault", regs, code, TRAP_BRKPT, |
| 110 | SIGSEGV) == NOTIFY_STOP) |
| 111 | return; |
| 112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | down_read(&mm->mmap_sem); |
| 114 | |
| 115 | vma = find_vma_prev(mm, address, &prev_vma); |
| 116 | if (!vma) |
| 117 | goto bad_area; |
| 118 | |
| 119 | /* find_vma_prev() returns vma such that address < vma->vm_end or NULL */ |
| 120 | if (address < vma->vm_start) |
| 121 | goto check_expansion; |
| 122 | |
| 123 | good_area: |
| 124 | code = SEGV_ACCERR; |
| 125 | |
| 126 | /* OK, we've got a good vm_area for this memory area. Check the access permissions: */ |
| 127 | |
| 128 | # define VM_READ_BIT 0 |
| 129 | # define VM_WRITE_BIT 1 |
| 130 | # define VM_EXEC_BIT 2 |
| 131 | |
| 132 | # if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) \ |
| 133 | || (1 << VM_EXEC_BIT) != VM_EXEC) |
| 134 | # error File is out of sync with <linux/mm.h>. Please update. |
| 135 | # endif |
| 136 | |
| 137 | mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT) |
| 138 | | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT) |
| 139 | | (((isr >> IA64_ISR_R_BIT) & 1UL) << VM_READ_BIT)); |
| 140 | |
| 141 | if ((vma->vm_flags & mask) != mask) |
| 142 | goto bad_area; |
| 143 | |
| 144 | survive: |
| 145 | /* |
| 146 | * If for any reason at all we couldn't handle the fault, make |
| 147 | * sure we exit gracefully rather than endlessly redo the |
| 148 | * fault. |
| 149 | */ |
| 150 | switch (handle_mm_fault(mm, vma, address, (mask & VM_WRITE) != 0)) { |
| 151 | case VM_FAULT_MINOR: |
| 152 | ++current->min_flt; |
| 153 | break; |
| 154 | case VM_FAULT_MAJOR: |
| 155 | ++current->maj_flt; |
| 156 | break; |
| 157 | case VM_FAULT_SIGBUS: |
| 158 | /* |
| 159 | * We ran out of memory, or some other thing happened |
| 160 | * to us that made us unable to handle the page fault |
| 161 | * gracefully. |
| 162 | */ |
| 163 | signal = SIGBUS; |
| 164 | goto bad_area; |
| 165 | case VM_FAULT_OOM: |
| 166 | goto out_of_memory; |
| 167 | default: |
| 168 | BUG(); |
| 169 | } |
| 170 | up_read(&mm->mmap_sem); |
| 171 | return; |
| 172 | |
| 173 | check_expansion: |
| 174 | if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) { |
| 175 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
| 176 | goto bad_area; |
| 177 | if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start) |
| 178 | || REGION_OFFSET(address) >= RGN_MAP_LIMIT) |
| 179 | goto bad_area; |
| 180 | if (expand_stack(vma, address)) |
| 181 | goto bad_area; |
| 182 | } else { |
| 183 | vma = prev_vma; |
| 184 | if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start) |
| 185 | || REGION_OFFSET(address) >= RGN_MAP_LIMIT) |
| 186 | goto bad_area; |
| 187 | if (expand_backing_store(vma, address)) |
| 188 | goto bad_area; |
| 189 | } |
| 190 | goto good_area; |
| 191 | |
| 192 | bad_area: |
| 193 | up_read(&mm->mmap_sem); |
| 194 | #ifdef CONFIG_VIRTUAL_MEM_MAP |
| 195 | bad_area_no_up: |
| 196 | #endif |
| 197 | if ((isr & IA64_ISR_SP) |
| 198 | || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) |
| 199 | { |
| 200 | /* |
| 201 | * This fault was due to a speculative load or lfetch.fault, set the "ed" |
| 202 | * bit in the psr to ensure forward progress. (Target register will get a |
| 203 | * NaT for ld.s, lfetch will be canceled.) |
| 204 | */ |
| 205 | ia64_psr(regs)->ed = 1; |
| 206 | return; |
| 207 | } |
| 208 | if (user_mode(regs)) { |
| 209 | si.si_signo = signal; |
| 210 | si.si_errno = 0; |
| 211 | si.si_code = code; |
| 212 | si.si_addr = (void __user *) address; |
| 213 | si.si_isr = isr; |
| 214 | si.si_flags = __ISR_VALID; |
| 215 | force_sig_info(signal, &si, current); |
| 216 | return; |
| 217 | } |
| 218 | |
| 219 | no_context: |
Tony Luck | f0a8d3c | 2005-04-25 13:22:44 -0700 | [diff] [blame] | 220 | if ((isr & IA64_ISR_SP) |
| 221 | || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) |
| 222 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | /* |
Tony Luck | f0a8d3c | 2005-04-25 13:22:44 -0700 | [diff] [blame] | 224 | * This fault was due to a speculative load or lfetch.fault, set the "ed" |
| 225 | * bit in the psr to ensure forward progress. (Target register will get a |
| 226 | * NaT for ld.s, lfetch will be canceled.) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | */ |
| 228 | ia64_psr(regs)->ed = 1; |
| 229 | return; |
| 230 | } |
| 231 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | /* |
| 233 | * Since we have no vma's for region 5, we might get here even if the address is |
| 234 | * valid, due to the VHPT walker inserting a non present translation that becomes |
| 235 | * stale. If that happens, the non present fault handler already purged the stale |
| 236 | * translation, which fixed the problem. So, we check to see if the translation is |
| 237 | * valid, and return if it is. |
| 238 | */ |
| 239 | if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address)) |
| 240 | return; |
| 241 | |
Kiyoshi Ueda | 63028aa | 2005-08-24 18:03:43 -0400 | [diff] [blame^] | 242 | if (ia64_done_with_exception(regs)) |
| 243 | return; |
| 244 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | /* |
| 246 | * Oops. The kernel tried to access some bad page. We'll have to terminate things |
| 247 | * with extreme prejudice. |
| 248 | */ |
| 249 | bust_spinlocks(1); |
| 250 | |
| 251 | if (address < PAGE_SIZE) |
| 252 | printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address); |
| 253 | else |
| 254 | printk(KERN_ALERT "Unable to handle kernel paging request at " |
| 255 | "virtual address %016lx\n", address); |
| 256 | die("Oops", regs, isr); |
| 257 | bust_spinlocks(0); |
| 258 | do_exit(SIGKILL); |
| 259 | return; |
| 260 | |
| 261 | out_of_memory: |
| 262 | up_read(&mm->mmap_sem); |
| 263 | if (current->pid == 1) { |
| 264 | yield(); |
| 265 | down_read(&mm->mmap_sem); |
| 266 | goto survive; |
| 267 | } |
| 268 | printk(KERN_CRIT "VM: killing process %s\n", current->comm); |
| 269 | if (user_mode(regs)) |
| 270 | do_exit(SIGKILL); |
| 271 | goto no_context; |
| 272 | } |