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 int 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() && kprobe_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 int fault; 84 85 /* mmap_sem is performance critical.... */ 86 prefetchw(&mm->mmap_sem); 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 105 106 /* 107 * This is to handle the kprobes on user space access instructions 108 */ 109 if (notify_page_fault(regs, TRAP_BRKPT)) 110 return; 111 112 down_read(&mm->mmap_sem); 113 114 vma = find_vma_prev(mm, address, &prev_vma); 115 if (!vma && !prev_vma ) 116 goto bad_area; 117 118 /* 119 * find_vma_prev() returns vma such that address < vma->vm_end or NULL 120 * 121 * May find no vma, but could be that the last vm area is the 122 * register backing store that needs to expand upwards, in 123 * this case vma will be null, but prev_vma will ne non-null 124 */ 125 if (( !vma && prev_vma ) || (address < vma->vm_start) ) 126 goto check_expansion; 127 128 good_area: 129 code = SEGV_ACCERR; 130 131 /* OK, we've got a good vm_area for this memory area. Check the access permissions: */ 132 133# define VM_READ_BIT 0 134# define VM_WRITE_BIT 1 135# define VM_EXEC_BIT 2 136 137# if (((1 << VM_READ_BIT) != VM_READ || (1 << VM_WRITE_BIT) != VM_WRITE) || (1 << \ 138 VM_EXEC_BIT) != VM_EXEC) 139# error File is out of sync with <linux/mm.h>. Please update. 140# endif 141 142 if (((isr >> IA64_ISR_R_BIT) & 1UL) && (!(vma->vm_flags & (VM_READ | VM_WRITE)))) 143 goto bad_area; 144 145 mask = ( (((isr >> IA64_ISR_X_BIT) & 1UL) << VM_EXEC_BIT) 146 | (((isr >> IA64_ISR_W_BIT) & 1UL) << VM_WRITE_BIT)); 147 148 if ((vma->vm_flags & mask) != mask) 149 goto bad_area; 150 151 /* 152 * If for any reason at all we couldn't handle the fault, make 153 * sure we exit gracefully rather than endlessly redo the 154 * fault. 155 */ 156 fault = handle_mm_fault(mm, vma, address, (mask & VM_WRITE) ? FAULT_FLAG_WRITE : 0); 157 if (unlikely(fault & VM_FAULT_ERROR)) { 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 if (fault & VM_FAULT_OOM) { 164 goto out_of_memory; 165 } else if (fault & VM_FAULT_SIGBUS) { 166 signal = SIGBUS; 167 goto bad_area; 168 } 169 BUG(); 170 } 171 if (fault & VM_FAULT_MAJOR) 172 current->maj_flt++; 173 else 174 current->min_flt++; 175 up_read(&mm->mmap_sem); 176 return; 177 178 check_expansion: 179 if (!(prev_vma && (prev_vma->vm_flags & VM_GROWSUP) && (address == prev_vma->vm_end))) { 180 if (!vma) 181 goto bad_area; 182 if (!(vma->vm_flags & VM_GROWSDOWN)) 183 goto bad_area; 184 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start) 185 || REGION_OFFSET(address) >= RGN_MAP_LIMIT) 186 goto bad_area; 187 if (expand_stack(vma, address)) 188 goto bad_area; 189 } else { 190 vma = prev_vma; 191 if (REGION_NUMBER(address) != REGION_NUMBER(vma->vm_start) 192 || REGION_OFFSET(address) >= RGN_MAP_LIMIT) 193 goto bad_area; 194 /* 195 * Since the register backing store is accessed sequentially, 196 * we disallow growing it by more than a page at a time. 197 */ 198 if (address > vma->vm_end + PAGE_SIZE - sizeof(long)) 199 goto bad_area; 200 if (expand_upwards(vma, address)) 201 goto bad_area; 202 } 203 goto good_area; 204 205 bad_area: 206 up_read(&mm->mmap_sem); 207#ifdef CONFIG_VIRTUAL_MEM_MAP 208 bad_area_no_up: 209#endif 210 if ((isr & IA64_ISR_SP) 211 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) 212 { 213 /* 214 * This fault was due to a speculative load or lfetch.fault, set the "ed" 215 * bit in the psr to ensure forward progress. (Target register will get a 216 * NaT for ld.s, lfetch will be canceled.) 217 */ 218 ia64_psr(regs)->ed = 1; 219 return; 220 } 221 if (user_mode(regs)) { 222 si.si_signo = signal; 223 si.si_errno = 0; 224 si.si_code = code; 225 si.si_addr = (void __user *) address; 226 si.si_isr = isr; 227 si.si_flags = __ISR_VALID; 228 force_sig_info(signal, &si, current); 229 return; 230 } 231 232 no_context: 233 if ((isr & IA64_ISR_SP) 234 || ((isr & IA64_ISR_NA) && (isr & IA64_ISR_CODE_MASK) == IA64_ISR_CODE_LFETCH)) 235 { 236 /* 237 * This fault was due to a speculative load or lfetch.fault, set the "ed" 238 * bit in the psr to ensure forward progress. (Target register will get a 239 * NaT for ld.s, lfetch will be canceled.) 240 */ 241 ia64_psr(regs)->ed = 1; 242 return; 243 } 244 245 /* 246 * Since we have no vma's for region 5, we might get here even if the address is 247 * valid, due to the VHPT walker inserting a non present translation that becomes 248 * stale. If that happens, the non present fault handler already purged the stale 249 * translation, which fixed the problem. So, we check to see if the translation is 250 * valid, and return if it is. 251 */ 252 if (REGION_NUMBER(address) == 5 && mapped_kernel_page_is_present(address)) 253 return; 254 255 if (ia64_done_with_exception(regs)) 256 return; 257 258 /* 259 * Oops. The kernel tried to access some bad page. We'll have to terminate things 260 * with extreme prejudice. 261 */ 262 bust_spinlocks(1); 263 264 if (address < PAGE_SIZE) 265 printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference (address %016lx)\n", address); 266 else 267 printk(KERN_ALERT "Unable to handle kernel paging request at " 268 "virtual address %016lx\n", address); 269 if (die("Oops", regs, isr)) 270 regs = NULL; 271 bust_spinlocks(0); 272 if (regs) 273 do_exit(SIGKILL); 274 return; 275 276 out_of_memory: 277 up_read(&mm->mmap_sem); 278 if (!user_mode(regs)) 279 goto no_context; 280 pagefault_out_of_memory(); 281} 282