1 /* 2 * Copyright (C) 2009 Wind River Systems Inc 3 * Implemented by fredrik.markstrom@gmail.com and ivarholmqvist@gmail.com 4 * 5 * based on arch/mips/mm/fault.c which is: 6 * 7 * Copyright (C) 1995-2000 Ralf Baechle 8 * 9 * This file is subject to the terms and conditions of the GNU General Public 10 * License. See the file "COPYING" in the main directory of this archive 11 * for more details. 12 */ 13 14 #include <linux/signal.h> 15 #include <linux/sched.h> 16 #include <linux/sched/debug.h> 17 #include <linux/interrupt.h> 18 #include <linux/kernel.h> 19 #include <linux/errno.h> 20 #include <linux/string.h> 21 #include <linux/types.h> 22 #include <linux/ptrace.h> 23 #include <linux/mman.h> 24 #include <linux/mm.h> 25 #include <linux/extable.h> 26 #include <linux/uaccess.h> 27 28 #include <asm/mmu_context.h> 29 #include <asm/traps.h> 30 31 #define EXC_SUPERV_INSN_ACCESS 9 /* Supervisor only instruction address */ 32 #define EXC_SUPERV_DATA_ACCESS 11 /* Supervisor only data address */ 33 #define EXC_X_PROTECTION_FAULT 13 /* TLB permission violation (x) */ 34 #define EXC_R_PROTECTION_FAULT 14 /* TLB permission violation (r) */ 35 #define EXC_W_PROTECTION_FAULT 15 /* TLB permission violation (w) */ 36 37 /* 38 * This routine handles page faults. It determines the address, 39 * and the problem, and then passes it off to one of the appropriate 40 * routines. 41 */ 42 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long cause, 43 unsigned long address) 44 { 45 struct vm_area_struct *vma = NULL; 46 struct task_struct *tsk = current; 47 struct mm_struct *mm = tsk->mm; 48 int code = SEGV_MAPERR; 49 vm_fault_t fault; 50 unsigned int flags = FAULT_FLAG_DEFAULT; 51 52 cause >>= 2; 53 54 /* Restart the instruction */ 55 regs->ea -= 4; 56 57 /* 58 * We fault-in kernel-space virtual memory on-demand. The 59 * 'reference' page table is init_mm.pgd. 60 * 61 * NOTE! We MUST NOT take any locks for this case. We may 62 * be in an interrupt or a critical region, and should 63 * only copy the information from the master page table, 64 * nothing more. 65 */ 66 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END)) { 67 if (user_mode(regs)) 68 goto bad_area_nosemaphore; 69 else 70 goto vmalloc_fault; 71 } 72 73 if (unlikely(address >= TASK_SIZE)) 74 goto bad_area_nosemaphore; 75 76 /* 77 * If we're in an interrupt or have no user 78 * context, we must not take the fault.. 79 */ 80 if (faulthandler_disabled() || !mm) 81 goto bad_area_nosemaphore; 82 83 if (user_mode(regs)) 84 flags |= FAULT_FLAG_USER; 85 86 if (!mmap_read_trylock(mm)) { 87 if (!user_mode(regs) && !search_exception_tables(regs->ea)) 88 goto bad_area_nosemaphore; 89 retry: 90 mmap_read_lock(mm); 91 } 92 93 vma = find_vma(mm, address); 94 if (!vma) 95 goto bad_area; 96 if (vma->vm_start <= address) 97 goto good_area; 98 if (!(vma->vm_flags & VM_GROWSDOWN)) 99 goto bad_area; 100 if (expand_stack(vma, address)) 101 goto bad_area; 102 /* 103 * Ok, we have a good vm_area for this memory access, so 104 * we can handle it.. 105 */ 106 good_area: 107 code = SEGV_ACCERR; 108 109 switch (cause) { 110 case EXC_SUPERV_INSN_ACCESS: 111 goto bad_area; 112 case EXC_SUPERV_DATA_ACCESS: 113 goto bad_area; 114 case EXC_X_PROTECTION_FAULT: 115 if (!(vma->vm_flags & VM_EXEC)) 116 goto bad_area; 117 break; 118 case EXC_R_PROTECTION_FAULT: 119 if (!(vma->vm_flags & VM_READ)) 120 goto bad_area; 121 break; 122 case EXC_W_PROTECTION_FAULT: 123 if (!(vma->vm_flags & VM_WRITE)) 124 goto bad_area; 125 flags = FAULT_FLAG_WRITE; 126 break; 127 } 128 129 /* 130 * If for any reason at all we couldn't handle the fault, 131 * make sure we exit gracefully rather than endlessly redo 132 * the fault. 133 */ 134 fault = handle_mm_fault(vma, address, flags); 135 136 if (fault_signal_pending(fault, regs)) 137 return; 138 139 if (unlikely(fault & VM_FAULT_ERROR)) { 140 if (fault & VM_FAULT_OOM) 141 goto out_of_memory; 142 else if (fault & VM_FAULT_SIGSEGV) 143 goto bad_area; 144 else if (fault & VM_FAULT_SIGBUS) 145 goto do_sigbus; 146 BUG(); 147 } 148 149 /* 150 * Major/minor page fault accounting is only done on the 151 * initial attempt. If we go through a retry, it is extremely 152 * likely that the page will be found in page cache at that point. 153 */ 154 if (flags & FAULT_FLAG_ALLOW_RETRY) { 155 if (fault & VM_FAULT_MAJOR) 156 current->maj_flt++; 157 else 158 current->min_flt++; 159 if (fault & VM_FAULT_RETRY) { 160 flags |= FAULT_FLAG_TRIED; 161 162 /* 163 * No need to mmap_read_unlock(mm) as we would 164 * have already released it in __lock_page_or_retry 165 * in mm/filemap.c. 166 */ 167 168 goto retry; 169 } 170 } 171 172 mmap_read_unlock(mm); 173 return; 174 175 /* 176 * Something tried to access memory that isn't in our memory map.. 177 * Fix it, but check if it's kernel or user first.. 178 */ 179 bad_area: 180 mmap_read_unlock(mm); 181 182 bad_area_nosemaphore: 183 /* User mode accesses just cause a SIGSEGV */ 184 if (user_mode(regs)) { 185 if (unhandled_signal(current, SIGSEGV) && printk_ratelimit()) { 186 pr_info("%s: unhandled page fault (%d) at 0x%08lx, " 187 "cause %ld\n", current->comm, SIGSEGV, address, cause); 188 show_regs(regs); 189 } 190 _exception(SIGSEGV, regs, code, address); 191 return; 192 } 193 194 no_context: 195 /* Are we prepared to handle this kernel fault? */ 196 if (fixup_exception(regs)) 197 return; 198 199 /* 200 * Oops. The kernel tried to access some bad page. We'll have to 201 * terminate things with extreme prejudice. 202 */ 203 bust_spinlocks(1); 204 205 pr_alert("Unable to handle kernel %s at virtual address %08lx", 206 address < PAGE_SIZE ? "NULL pointer dereference" : 207 "paging request", address); 208 pr_alert("ea = %08lx, ra = %08lx, cause = %ld\n", regs->ea, regs->ra, 209 cause); 210 panic("Oops"); 211 return; 212 213 /* 214 * We ran out of memory, or some other thing happened to us that made 215 * us unable to handle the page fault gracefully. 216 */ 217 out_of_memory: 218 mmap_read_unlock(mm); 219 if (!user_mode(regs)) 220 goto no_context; 221 pagefault_out_of_memory(); 222 return; 223 224 do_sigbus: 225 mmap_read_unlock(mm); 226 227 /* Kernel mode? Handle exceptions or die */ 228 if (!user_mode(regs)) 229 goto no_context; 230 231 _exception(SIGBUS, regs, BUS_ADRERR, address); 232 return; 233 234 vmalloc_fault: 235 { 236 /* 237 * Synchronize this task's top level page-table 238 * with the 'reference' page table. 239 * 240 * Do _not_ use "tsk" here. We might be inside 241 * an interrupt in the middle of a task switch.. 242 */ 243 int offset = pgd_index(address); 244 pgd_t *pgd, *pgd_k; 245 p4d_t *p4d, *p4d_k; 246 pud_t *pud, *pud_k; 247 pmd_t *pmd, *pmd_k; 248 pte_t *pte_k; 249 250 pgd = pgd_current + offset; 251 pgd_k = init_mm.pgd + offset; 252 253 if (!pgd_present(*pgd_k)) 254 goto no_context; 255 set_pgd(pgd, *pgd_k); 256 257 p4d = p4d_offset(pgd, address); 258 p4d_k = p4d_offset(pgd_k, address); 259 if (!p4d_present(*p4d_k)) 260 goto no_context; 261 pud = pud_offset(p4d, address); 262 pud_k = pud_offset(p4d_k, address); 263 if (!pud_present(*pud_k)) 264 goto no_context; 265 pmd = pmd_offset(pud, address); 266 pmd_k = pmd_offset(pud_k, address); 267 if (!pmd_present(*pmd_k)) 268 goto no_context; 269 set_pmd(pmd, *pmd_k); 270 271 pte_k = pte_offset_kernel(pmd_k, address); 272 if (!pte_present(*pte_k)) 273 goto no_context; 274 275 flush_tlb_kernel_page(address); 276 return; 277 } 278 } 279