1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Copyright (C) 2009 Sunplus Core Technology Co., Ltd. 4 * Lennox Wu <lennox.wu@sunplusct.com> 5 * Chen Liqin <liqin.chen@sunplusct.com> 6 * Copyright (C) 2012 Regents of the University of California 7 */ 8 9 10 #include <linux/mm.h> 11 #include <linux/kernel.h> 12 #include <linux/interrupt.h> 13 #include <linux/perf_event.h> 14 #include <linux/signal.h> 15 #include <linux/uaccess.h> 16 17 #include <asm/ptrace.h> 18 #include <asm/tlbflush.h> 19 20 #include "../kernel/head.h" 21 22 static inline void no_context(struct pt_regs *regs, unsigned long addr) 23 { 24 /* Are we prepared to handle this kernel fault? */ 25 if (fixup_exception(regs)) 26 return; 27 28 /* 29 * Oops. The kernel tried to access some bad page. We'll have to 30 * terminate things with extreme prejudice. 31 */ 32 bust_spinlocks(1); 33 pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", 34 (addr < PAGE_SIZE) ? "NULL pointer dereference" : 35 "paging request", addr); 36 die(regs, "Oops"); 37 do_exit(SIGKILL); 38 } 39 40 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault) 41 { 42 if (fault & VM_FAULT_OOM) { 43 /* 44 * We ran out of memory, call the OOM killer, and return the userspace 45 * (which will retry the fault, or kill us if we got oom-killed). 46 */ 47 if (!user_mode(regs)) { 48 no_context(regs, addr); 49 return; 50 } 51 pagefault_out_of_memory(); 52 return; 53 } else if (fault & VM_FAULT_SIGBUS) { 54 /* Kernel mode? Handle exceptions or die */ 55 if (!user_mode(regs)) { 56 no_context(regs, addr); 57 return; 58 } 59 do_trap(regs, SIGBUS, BUS_ADRERR, addr); 60 return; 61 } 62 BUG(); 63 } 64 65 static inline void bad_area(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr) 66 { 67 /* 68 * Something tried to access memory that isn't in our memory map. 69 * Fix it, but check if it's kernel or user first. 70 */ 71 mmap_read_unlock(mm); 72 /* User mode accesses just cause a SIGSEGV */ 73 if (user_mode(regs)) { 74 do_trap(regs, SIGSEGV, code, addr); 75 return; 76 } 77 78 no_context(regs, addr); 79 } 80 81 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr) 82 { 83 pgd_t *pgd, *pgd_k; 84 pud_t *pud, *pud_k; 85 p4d_t *p4d, *p4d_k; 86 pmd_t *pmd, *pmd_k; 87 pte_t *pte_k; 88 int index; 89 90 /* User mode accesses just cause a SIGSEGV */ 91 if (user_mode(regs)) 92 return do_trap(regs, SIGSEGV, code, addr); 93 94 /* 95 * Synchronize this task's top level page-table 96 * with the 'reference' page table. 97 * 98 * Do _not_ use "tsk->active_mm->pgd" here. 99 * We might be inside an interrupt in the middle 100 * of a task switch. 101 */ 102 index = pgd_index(addr); 103 pgd = (pgd_t *)pfn_to_virt(csr_read(CSR_SATP)) + index; 104 pgd_k = init_mm.pgd + index; 105 106 if (!pgd_present(*pgd_k)) { 107 no_context(regs, addr); 108 return; 109 } 110 set_pgd(pgd, *pgd_k); 111 112 p4d = p4d_offset(pgd, addr); 113 p4d_k = p4d_offset(pgd_k, addr); 114 if (!p4d_present(*p4d_k)) { 115 no_context(regs, addr); 116 return; 117 } 118 119 pud = pud_offset(p4d, addr); 120 pud_k = pud_offset(p4d_k, addr); 121 if (!pud_present(*pud_k)) { 122 no_context(regs, addr); 123 return; 124 } 125 126 /* 127 * Since the vmalloc area is global, it is unnecessary 128 * to copy individual PTEs 129 */ 130 pmd = pmd_offset(pud, addr); 131 pmd_k = pmd_offset(pud_k, addr); 132 if (!pmd_present(*pmd_k)) { 133 no_context(regs, addr); 134 return; 135 } 136 set_pmd(pmd, *pmd_k); 137 138 /* 139 * Make sure the actual PTE exists as well to 140 * catch kernel vmalloc-area accesses to non-mapped 141 * addresses. If we don't do this, this will just 142 * silently loop forever. 143 */ 144 pte_k = pte_offset_kernel(pmd_k, addr); 145 if (!pte_present(*pte_k)) { 146 no_context(regs, addr); 147 return; 148 } 149 150 /* 151 * The kernel assumes that TLBs don't cache invalid 152 * entries, but in RISC-V, SFENCE.VMA specifies an 153 * ordering constraint, not a cache flush; it is 154 * necessary even after writing invalid entries. 155 */ 156 local_flush_tlb_page(addr); 157 } 158 159 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma) 160 { 161 switch (cause) { 162 case EXC_INST_PAGE_FAULT: 163 if (!(vma->vm_flags & VM_EXEC)) { 164 return true; 165 } 166 break; 167 case EXC_LOAD_PAGE_FAULT: 168 if (!(vma->vm_flags & VM_READ)) { 169 return true; 170 } 171 break; 172 case EXC_STORE_PAGE_FAULT: 173 if (!(vma->vm_flags & VM_WRITE)) { 174 return true; 175 } 176 break; 177 default: 178 panic("%s: unhandled cause %lu", __func__, cause); 179 } 180 return false; 181 } 182 183 /* 184 * This routine handles page faults. It determines the address and the 185 * problem, and then passes it off to one of the appropriate routines. 186 */ 187 asmlinkage void do_page_fault(struct pt_regs *regs) 188 { 189 struct task_struct *tsk; 190 struct vm_area_struct *vma; 191 struct mm_struct *mm; 192 unsigned long addr, cause; 193 unsigned int flags = FAULT_FLAG_DEFAULT; 194 int code = SEGV_MAPERR; 195 vm_fault_t fault; 196 197 cause = regs->cause; 198 addr = regs->badaddr; 199 200 tsk = current; 201 mm = tsk->mm; 202 203 /* 204 * Fault-in kernel-space virtual memory on-demand. 205 * The 'reference' page table is init_mm.pgd. 206 * 207 * NOTE! We MUST NOT take any locks for this case. We may 208 * be in an interrupt or a critical region, and should 209 * only copy the information from the master page table, 210 * nothing more. 211 */ 212 if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END))) { 213 vmalloc_fault(regs, code, addr); 214 return; 215 } 216 217 /* Enable interrupts if they were enabled in the parent context. */ 218 if (likely(regs->status & SR_PIE)) 219 local_irq_enable(); 220 221 /* 222 * If we're in an interrupt, have no user context, or are running 223 * in an atomic region, then we must not take the fault. 224 */ 225 if (unlikely(faulthandler_disabled() || !mm)) { 226 no_context(regs, addr); 227 return; 228 } 229 230 if (user_mode(regs)) 231 flags |= FAULT_FLAG_USER; 232 233 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr); 234 235 if (cause == EXC_STORE_PAGE_FAULT) 236 flags |= FAULT_FLAG_WRITE; 237 else if (cause == EXC_INST_PAGE_FAULT) 238 flags |= FAULT_FLAG_INSTRUCTION; 239 retry: 240 mmap_read_lock(mm); 241 vma = find_vma(mm, addr); 242 if (unlikely(!vma)) { 243 bad_area(regs, mm, code, addr); 244 return; 245 } 246 if (likely(vma->vm_start <= addr)) 247 goto good_area; 248 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 249 bad_area(regs, mm, code, addr); 250 return; 251 } 252 if (unlikely(expand_stack(vma, addr))) { 253 bad_area(regs, mm, code, addr); 254 return; 255 } 256 257 /* 258 * Ok, we have a good vm_area for this memory access, so 259 * we can handle it. 260 */ 261 good_area: 262 code = SEGV_ACCERR; 263 264 if (unlikely(access_error(cause, vma))) { 265 bad_area(regs, mm, code, addr); 266 return; 267 } 268 269 /* 270 * If for any reason at all we could not handle the fault, 271 * make sure we exit gracefully rather than endlessly redo 272 * the fault. 273 */ 274 fault = handle_mm_fault(vma, addr, flags, regs); 275 276 /* 277 * If we need to retry but a fatal signal is pending, handle the 278 * signal first. We do not need to release the mmap_lock because it 279 * would already be released in __lock_page_or_retry in mm/filemap.c. 280 */ 281 if (fault_signal_pending(fault, regs)) 282 return; 283 284 if (unlikely((fault & VM_FAULT_RETRY) && (flags & FAULT_FLAG_ALLOW_RETRY))) { 285 flags |= FAULT_FLAG_TRIED; 286 287 /* 288 * No need to mmap_read_unlock(mm) as we would 289 * have already released it in __lock_page_or_retry 290 * in mm/filemap.c. 291 */ 292 goto retry; 293 } 294 295 mmap_read_unlock(mm); 296 297 if (unlikely(fault & VM_FAULT_ERROR)) { 298 mm_fault_error(regs, addr, fault); 299 return; 300 } 301 return; 302 } 303