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/interrupt.h> 11#include <linux/kprobes.h> 12#include <linux/kdebug.h> 13 14#include <asm/pgtable.h> 15#include <asm/processor.h> 16#include <asm/system.h> 17#include <asm/uaccess.h> 18 19extern void die (char *, struct pt_regs *, long); 20 21#ifdef CONFIG_KPROBES 22static inline int notify_page_fault(struct pt_regs *regs, int trap) 23{ 24 int ret = 0; 25 26 if (!user_mode(regs)) { 27 /* kprobe_running() needs smp_processor_id() */ 28 preempt_disable(); 29 if (kprobe_running() && kprobes_fault_handler(regs, trap)) 30 ret = 1; 31 preempt_enable(); 32 } 33 34 return ret; 35} 36#else 37static inline int notify_page_fault(struct pt_regs *regs, int trap) 38{ 39 return 0; 40} 41#endif 42 43/* 44 * Return TRUE if ADDRESS points at a page in the kernel's mapped segment 45 * (inside region 5, on ia64) and that page is present. 46 */ 47static int 48mapped_kernel_page_is_present (unsigned long address) 49{ 50 pgd_t *pgd; 51 pud_t *pud; 52 pmd_t *pmd; 53 pte_t *ptep, pte; 54 55 pgd = pgd_offset_k(address); 56 if (pgd_none(*pgd) || pgd_bad(*pgd)) 57 return 0; 58 59 pud = pud_offset(pgd, address); 60 if (pud_none(*pud) || pud_bad(*pud)) 61 return 0; 62 63 pmd = pmd_offset(pud, address); 64 if (pmd_none(*pmd) || pmd_bad(*pmd)) 65 return 0; 66 67 ptep = pte_offset_kernel(pmd, address); 68 if (!ptep) 69 return 0; 70 71 pte = *ptep; 72 return pte_present(pte); 73} 74 75void __kprobes 76ia64_do_page_fault (unsigned long address, unsigned long isr, struct pt_regs *regs) 77{ 78 int signal = SIGSEGV, code = SEGV_MAPERR; 79 struct vm_area_struct *vma, *prev_vma; 80 struct mm_struct *mm = current->mm; 81 struct siginfo si; 82 unsigned long mask; 83 84 /* mmap_sem is performance critical.... */ 85 prefetchw(&mm->mmap_sem); 86 87 /* 88 * If we're in an interrupt or have no user context, we must not take the fault.. 89 */ 90 if (in_atomic() || !mm) 91 goto no_context; 92 93#ifdef CONFIG_VIRTUAL_MEM_MAP 94 /* 95 * If fault is in region 5 and we are in the kernel, we may already 96 * have the mmap_sem (pfn_valid macro is called during mmap). There 97 * is no vma for region 5 addr's anyway, so skip getting the semaphore 98 * and go directly to the exception handling code. 99 */ 100 101 if ((REGION_NUMBER(address) == 5) && !user_mode(regs)) 102 goto bad_area_no_up; 103#endif 104 105 /* 106 * This is to handle the kprobes on user space access instructions 107 */ 108 if (notify_page_fault(regs, TRAP_BRKPT)) 109 return; 110 111 down_read(&mm->mmap_sem); 112 113 vma = find_vma_prev(mm, address, &prev_vma); 114 if (!vma) 115 goto bad_area; 116 117 /* find_vma_prev() returns vma such that address < vma->vm_end or NULL */ 118 if (address < vma->vm_start) 119 goto check_expansion; 120 121 good_area: 122 code = SEGV_ACCERR; 123 124 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */ 125 126# define VM_READ_BIT 0 127# define VM_WRITE_BIT 1 128# define VM_EXEC_BIT 2 129 130# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) || (1 << \ 131 VM_EXEC_BIT) != VM_EXEC) 132# error File is out of sync with <linux/mm.h>. Please update. 133# endif 134 135 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE)))) 136 goto bad_area; 137 138 mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT) 139 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_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 /* 188 * Since the register backing store is accessed sequentially, 189 * we disallow growing it by more than a page at a time. 190 */ 191 if (address > vma->vm_end + PAGE_SIZE - sizeof(long)) 192 goto bad_area; 193 if (expand_upwards(vma, address)) 194 goto bad_area; 195 } 196 goto good_area; 197 198 bad_area: 199 up_read(&mm->mmap_sem); 200#ifdef CONFIG_VIRTUAL_MEM_MAP 201 bad_area_no_up: 202#endif 203 if ((isr & IA64_ISR_SP) 204 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) 205 { 206 /* 207 * This fault was due to a speculative load or lfetch.fault, set the "ed" 208 * bit in the psr to ensure forward progress. (Target register will get a 209 * NaT for ld.s, lfetch will be canceled.) 210 */ 211 ia64_psr(regs)->ed = 1; 212 return; 213 } 214 if (user_mode(regs)) { 215 si.si_signo = signal; 216 si.si_errno = 0; 217 si.si_code = code; 218 si.si_addr = (void __user *) address; 219 si.si_isr = isr; 220 si.si_flags = __ISR_VALID; 221 force_sig_info(signal, &si, current); 222 return; 223 } 224 225 no_context: 226 if ((isr & IA64_ISR_SP) 227 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) 228 { 229 /* 230 * This fault was due to a speculative load or lfetch.fault, set the "ed" 231 * bit in the psr to ensure forward progress. (Target register will get a 232 * NaT for ld.s, lfetch will be canceled.) 233 */ 234 ia64_psr(regs)->ed = 1; 235 return; 236 } 237 238 /* 239 * Since we have no vma's for region 5, we might get here even if the address is 240 * valid, due to the VHPT walker inserting a non present translation that becomes 241 * stale. If that happens, the non present fault handler already purged the stale 242 * translation, which fixed the problem. So, we check to see if the translation is 243 * valid, and return if it is. 244 */ 245 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address)) 246 return; 247 248 if (ia64_done_with_exception(regs)) 249 return; 250 251 /* 252 * Oops. The kernel tried to access some bad page. We'll have to terminate things 253 * with extreme prejudice. 254 */ 255 bust_spinlocks(1); 256 257 if (address < PAGE_SIZE) 258 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address); 259 else 260 printk(KERN_ALERT "Unable to handle kernel paging request at " 261 "virtual address %016lx\n", address); 262 die("Oops", regs, isr); 263 bust_spinlocks(0); 264 do_exit(SIGKILL); 265 return; 266 267 out_of_memory: 268 up_read(&mm->mmap_sem); 269 if (is_init(current)) { 270 yield(); 271 down_read(&mm->mmap_sem); 272 goto survive; 273 } 274 printk(KERN_CRIT "VM: killing process %s\n", current->comm); 275 if (user_mode(regs)) 276 do_exit(SIGKILL); 277 goto no_context; 278} 279