| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * OpenRISC fault.c |
| 4 | * |
| 5 | * Linux architectural port borrowing liberally from similar works of |
| 6 | * others. All original copyrights apply as per the original source |
| 7 | * declaration. |
| 8 | * |
| 9 | * Modifications for the OpenRISC architecture: |
| 10 | * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com> |
| 11 | * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se> |
| 12 | */ |
| 13 | |
| 14 | #include <linux/mm.h> |
| 15 | #include <linux/interrupt.h> |
| 16 | #include <linux/extable.h> |
| 17 | #include <linux/sched/signal.h> |
| 18 | |
| 19 | #include <linux/uaccess.h> |
| 20 | #include <asm/siginfo.h> |
| 21 | #include <asm/signal.h> |
| 22 | |
| 23 | #define NUM_TLB_ENTRIES 64 |
| 24 | #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1)) |
| 25 | |
| 26 | unsigned long pte_misses; /* updated by do_page_fault() */ |
| 27 | unsigned long pte_errors; /* updated by do_page_fault() */ |
| 28 | |
| 29 | /* __PHX__ :: - check the vmalloc_fault in do_page_fault() |
| 30 | * - also look into include/asm-or32/mmu_context.h |
| 31 | */ |
| 32 | volatile pgd_t *current_pgd[NR_CPUS]; |
| 33 | |
| 34 | extern void die(char *, struct pt_regs *, long); |
| 35 | |
| 36 | /* |
| 37 | * This routine handles page faults. It determines the address, |
| 38 | * and the problem, and then passes it off to one of the appropriate |
| 39 | * routines. |
| 40 | * |
| 41 | * If this routine detects a bad access, it returns 1, otherwise it |
| 42 | * returns 0. |
| 43 | */ |
| 44 | |
| 45 | asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address, |
| 46 | unsigned long vector, int write_acc) |
| 47 | { |
| 48 | struct task_struct *tsk; |
| 49 | struct mm_struct *mm; |
| 50 | struct vm_area_struct *vma; |
| 51 | int si_code; |
| 52 | vm_fault_t fault; |
| 53 | unsigned int flags = FAULT_FLAG_DEFAULT; |
| 54 | |
| 55 | tsk = current; |
| 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 | * NOTE2: This is done so that, when updating the vmalloc |
| 67 | * mappings we don't have to walk all processes pgdirs and |
| 68 | * add the high mappings all at once. Instead we do it as they |
| 69 | * are used. However vmalloc'ed page entries have the PAGE_GLOBAL |
| 70 | * bit set so sometimes the TLB can use a lingering entry. |
| 71 | * |
| 72 | * This verifies that the fault happens in kernel space |
| 73 | * and that the fault was not a protection error. |
| 74 | */ |
| 75 | |
| 76 | if (address >= VMALLOC_START && |
| 77 | (vector != 0x300 && vector != 0x400) && |
| 78 | !user_mode(regs)) |
| 79 | goto vmalloc_fault; |
| 80 | |
| 81 | /* If exceptions were enabled, we can reenable them here */ |
| 82 | if (user_mode(regs)) { |
| 83 | /* Exception was in userspace: reenable interrupts */ |
| 84 | local_irq_enable(); |
| 85 | flags |= FAULT_FLAG_USER; |
| 86 | } else { |
| 87 | /* If exception was in a syscall, then IRQ's may have |
| 88 | * been enabled or disabled. If they were enabled, |
| 89 | * reenable them. |
| 90 | */ |
| 91 | if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE)) |
| 92 | local_irq_enable(); |
| 93 | } |
| 94 | |
| 95 | mm = tsk->mm; |
| 96 | si_code = SEGV_MAPERR; |
| 97 | |
| 98 | /* |
| 99 | * If we're in an interrupt or have no user |
| 100 | * context, we must not take the fault.. |
| 101 | */ |
| 102 | |
| 103 | if (in_interrupt() || !mm) |
| 104 | goto no_context; |
| 105 | |
| 106 | retry: |
| 107 | mmap_read_lock(mm); |
| 108 | vma = find_vma(mm, address); |
| 109 | |
| 110 | if (!vma) |
| 111 | goto bad_area; |
| 112 | |
| 113 | if (vma->vm_start <= address) |
| 114 | goto good_area; |
| 115 | |
| 116 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
| 117 | goto bad_area; |
| 118 | |
| 119 | if (user_mode(regs)) { |
| 120 | /* |
| 121 | * accessing the stack below usp is always a bug. |
| 122 | * we get page-aligned addresses so we can only check |
| 123 | * if we're within a page from usp, but that might be |
| 124 | * enough to catch brutal errors at least. |
| 125 | */ |
| 126 | if (address + PAGE_SIZE < regs->sp) |
| 127 | goto bad_area; |
| 128 | } |
| 129 | if (expand_stack(vma, address)) |
| 130 | goto bad_area; |
| 131 | |
| 132 | /* |
| 133 | * Ok, we have a good vm_area for this memory access, so |
| 134 | * we can handle it.. |
| 135 | */ |
| 136 | |
| 137 | good_area: |
| 138 | si_code = SEGV_ACCERR; |
| 139 | |
| 140 | /* first do some preliminary protection checks */ |
| 141 | |
| 142 | if (write_acc) { |
| 143 | if (!(vma->vm_flags & VM_WRITE)) |
| 144 | goto bad_area; |
| 145 | flags |= FAULT_FLAG_WRITE; |
| 146 | } else { |
| 147 | /* not present */ |
| 148 | if (!(vma->vm_flags & (VM_READ | VM_EXEC))) |
| 149 | goto bad_area; |
| 150 | } |
| 151 | |
| 152 | /* are we trying to execute nonexecutable area */ |
| 153 | if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC)) |
| 154 | goto bad_area; |
| 155 | |
| 156 | /* |
| 157 | * If for any reason at all we couldn't handle the fault, |
| 158 | * make sure we exit gracefully rather than endlessly redo |
| 159 | * the fault. |
| 160 | */ |
| 161 | |
| 162 | fault = handle_mm_fault(vma, address, flags); |
| 163 | |
| 164 | if (fault_signal_pending(fault, regs)) |
| 165 | return; |
| 166 | |
| 167 | if (unlikely(fault & VM_FAULT_ERROR)) { |
| 168 | if (fault & VM_FAULT_OOM) |
| 169 | goto out_of_memory; |
| 170 | else if (fault & VM_FAULT_SIGSEGV) |
| 171 | goto bad_area; |
| 172 | else if (fault & VM_FAULT_SIGBUS) |
| 173 | goto do_sigbus; |
| 174 | BUG(); |
| 175 | } |
| 176 | |
| 177 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
| 178 | /*RGD modeled on Cris */ |
| 179 | if (fault & VM_FAULT_MAJOR) |
| 180 | tsk->maj_flt++; |
| 181 | else |
| 182 | tsk->min_flt++; |
| 183 | if (fault & VM_FAULT_RETRY) { |
| 184 | flags |= FAULT_FLAG_TRIED; |
| 185 | |
| 186 | /* No need to mmap_read_unlock(mm) as we would |
| 187 | * have already released it in __lock_page_or_retry |
| 188 | * in mm/filemap.c. |
| 189 | */ |
| 190 | |
| 191 | goto retry; |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | mmap_read_unlock(mm); |
| 196 | return; |
| 197 | |
| 198 | /* |
| 199 | * Something tried to access memory that isn't in our memory map.. |
| 200 | * Fix it, but check if it's kernel or user first.. |
| 201 | */ |
| 202 | |
| 203 | bad_area: |
| 204 | mmap_read_unlock(mm); |
| 205 | |
| 206 | bad_area_nosemaphore: |
| 207 | |
| 208 | /* User mode accesses just cause a SIGSEGV */ |
| 209 | |
| 210 | if (user_mode(regs)) { |
| 211 | force_sig_fault(SIGSEGV, si_code, (void __user *)address); |
| 212 | return; |
| 213 | } |
| 214 | |
| 215 | no_context: |
| 216 | |
| 217 | /* Are we prepared to handle this kernel fault? |
| 218 | * |
| 219 | * (The kernel has valid exception-points in the source |
| 220 | * when it acesses user-memory. When it fails in one |
| 221 | * of those points, we find it in a table and do a jump |
| 222 | * to some fixup code that loads an appropriate error |
| 223 | * code) |
| 224 | */ |
| 225 | |
| 226 | { |
| 227 | const struct exception_table_entry *entry; |
| 228 | |
| 229 | __asm__ __volatile__("l.nop 42"); |
| 230 | |
| 231 | if ((entry = search_exception_tables(regs->pc)) != NULL) { |
| 232 | /* Adjust the instruction pointer in the stackframe */ |
| 233 | regs->pc = entry->fixup; |
| 234 | return; |
| 235 | } |
| 236 | } |
| 237 | |
| 238 | /* |
| 239 | * Oops. The kernel tried to access some bad page. We'll have to |
| 240 | * terminate things with extreme prejudice. |
| 241 | */ |
| 242 | |
| 243 | if ((unsigned long)(address) < PAGE_SIZE) |
| 244 | printk(KERN_ALERT |
| 245 | "Unable to handle kernel NULL pointer dereference"); |
| 246 | else |
| 247 | printk(KERN_ALERT "Unable to handle kernel access"); |
| 248 | printk(" at virtual address 0x%08lx\n", address); |
| 249 | |
| 250 | die("Oops", regs, write_acc); |
| 251 | |
| 252 | do_exit(SIGKILL); |
| 253 | |
| 254 | /* |
| 255 | * We ran out of memory, or some other thing happened to us that made |
| 256 | * us unable to handle the page fault gracefully. |
| 257 | */ |
| 258 | |
| 259 | out_of_memory: |
| 260 | __asm__ __volatile__("l.nop 42"); |
| 261 | __asm__ __volatile__("l.nop 1"); |
| 262 | |
| 263 | mmap_read_unlock(mm); |
| 264 | if (!user_mode(regs)) |
| 265 | goto no_context; |
| 266 | pagefault_out_of_memory(); |
| 267 | return; |
| 268 | |
| 269 | do_sigbus: |
| 270 | mmap_read_unlock(mm); |
| 271 | |
| 272 | /* |
| 273 | * Send a sigbus, regardless of whether we were in kernel |
| 274 | * or user mode. |
| 275 | */ |
| 276 | force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); |
| 277 | |
| 278 | /* Kernel mode? Handle exceptions or die */ |
| 279 | if (!user_mode(regs)) |
| 280 | goto no_context; |
| 281 | return; |
| 282 | |
| 283 | vmalloc_fault: |
| 284 | { |
| 285 | /* |
| 286 | * Synchronize this task's top level page-table |
| 287 | * with the 'reference' page table. |
| 288 | * |
| 289 | * Use current_pgd instead of tsk->active_mm->pgd |
| 290 | * since the latter might be unavailable if this |
| 291 | * code is executed in a misfortunately run irq |
| 292 | * (like inside schedule() between switch_mm and |
| 293 | * switch_to...). |
| 294 | */ |
| 295 | |
| 296 | int offset = pgd_index(address); |
| 297 | pgd_t *pgd, *pgd_k; |
| 298 | p4d_t *p4d, *p4d_k; |
| 299 | pud_t *pud, *pud_k; |
| 300 | pmd_t *pmd, *pmd_k; |
| 301 | pte_t *pte_k; |
| 302 | |
| 303 | /* |
| 304 | phx_warn("do_page_fault(): vmalloc_fault will not work, " |
| 305 | "since current_pgd assign a proper value somewhere\n" |
| 306 | "anyhow we don't need this at the moment\n"); |
| 307 | |
| 308 | phx_mmu("vmalloc_fault"); |
| 309 | */ |
| 310 | pgd = (pgd_t *)current_pgd[smp_processor_id()] + offset; |
| 311 | pgd_k = init_mm.pgd + offset; |
| 312 | |
| 313 | /* Since we're two-level, we don't need to do both |
| 314 | * set_pgd and set_pmd (they do the same thing). If |
| 315 | * we go three-level at some point, do the right thing |
| 316 | * with pgd_present and set_pgd here. |
| 317 | * |
| 318 | * Also, since the vmalloc area is global, we don't |
| 319 | * need to copy individual PTE's, it is enough to |
| 320 | * copy the pgd pointer into the pte page of the |
| 321 | * root task. If that is there, we'll find our pte if |
| 322 | * it exists. |
| 323 | */ |
| 324 | |
| 325 | p4d = p4d_offset(pgd, address); |
| 326 | p4d_k = p4d_offset(pgd_k, address); |
| 327 | if (!p4d_present(*p4d_k)) |
| 328 | goto no_context; |
| 329 | |
| 330 | pud = pud_offset(p4d, address); |
| 331 | pud_k = pud_offset(p4d_k, address); |
| 332 | if (!pud_present(*pud_k)) |
| 333 | goto no_context; |
| 334 | |
| 335 | pmd = pmd_offset(pud, address); |
| 336 | pmd_k = pmd_offset(pud_k, address); |
| 337 | |
| 338 | if (!pmd_present(*pmd_k)) |
| 339 | goto bad_area_nosemaphore; |
| 340 | |
| 341 | set_pmd(pmd, *pmd_k); |
| 342 | |
| 343 | /* Make sure the actual PTE exists as well to |
| 344 | * catch kernel vmalloc-area accesses to non-mapped |
| 345 | * addresses. If we don't do this, this will just |
| 346 | * silently loop forever. |
| 347 | */ |
| 348 | |
| 349 | pte_k = pte_offset_kernel(pmd_k, address); |
| 350 | if (!pte_present(*pte_k)) |
| 351 | goto no_context; |
| 352 | |
| 353 | return; |
| 354 | } |
| 355 | } |