arch/parisc/mm/fault.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  *
   7  * Copyright (C) 1995, 1996, 1997, 1998 by Ralf Baechle
   8  * Copyright 1999 SuSE GmbH (Philipp Rumpf, prumpf@tux.org)
   9  * Copyright 1999 Hewlett Packard Co.
  10  *
  11  */
  12
  13 #include <linux/mm.h>
  14 #include <linux/ptrace.h>
  15 #include <linux/sched.h>
  16 #include <linux/sched/debug.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/extable.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/hugetlb.h>
  21 #include <linux/perf_event.h>
  22
  23 #include <asm/traps.h>
  24
  25 #define DEBUG_NATLB 0
  26
  27 /* Various important other fields */
  28 #define bit22set(x)             (x & 0x00000200)
  29 #define bits23_25set(x)         (x & 0x000001c0)
  30 #define isGraphicsFlushRead(x)  ((x & 0xfc003fdf) == 0x04001a80)
  31                                 /* extended opcode is 0x6a */
  32
  33 #define BITSSET         0x1c0   /* for identifying LDCW */
  34
  35
  36 int show_unhandled_signals = 1;
  37
  38 /*
  39  * parisc_acctyp(unsigned int inst) --
  40  *    Given a PA-RISC memory access instruction, determine if the
  41  *    instruction would perform a memory read or memory write
  42  *    operation.
  43  *
  44  *    This function assumes that the given instruction is a memory access
  45  *    instruction (i.e. you should really only call it if you know that
  46  *    the instruction has generated some sort of a memory access fault).
  47  *
  48  * Returns:
  49  *   VM_READ  if read operation
  50  *   VM_WRITE if write operation
  51  *   VM_EXEC  if execute operation
  52  */
  53 unsigned long
  54 parisc_acctyp(unsigned long code, unsigned int inst)
  55 {
  56         if (code == 6 || code == 16)
  57             return VM_EXEC;
  58
  59         switch (inst & 0xf0000000) {
  60         case 0x40000000: /* load */
  61         case 0x50000000: /* new load */
  62                 return VM_READ;
  63
  64         case 0x60000000: /* store */
  65         case 0x70000000: /* new store */
  66                 return VM_WRITE;
  67
  68         case 0x20000000: /* coproc */
  69         case 0x30000000: /* coproc2 */
  70                 if (bit22set(inst))
  71                         return VM_WRITE;
  72                 fallthrough;
  73
  74         case 0x0: /* indexed/memory management */
  75                 if (bit22set(inst)) {
  76                         /*
  77                          * Check for the 'Graphics Flush Read' instruction.
  78                          * It resembles an FDC instruction, except for bits
  79                          * 20 and 21. Any combination other than zero will
  80                          * utilize the block mover functionality on some
  81                          * older PA-RISC platforms.  The case where a block
  82                          * move is performed from VM to graphics IO space
  83                          * should be treated as a READ.
  84                          *
  85                          * The significance of bits 20,21 in the FDC
  86                          * instruction is:
  87                          *
  88                          *   00  Flush data cache (normal instruction behavior)
  89                          *   01  Graphics flush write  (IO space -> VM)
  90                          *   10  Graphics flush read   (VM -> IO space)
  91                          *   11  Graphics flush read/write (VM <-> IO space)
  92                          */
  93                         if (isGraphicsFlushRead(inst))
  94                                 return VM_READ;
  95                         return VM_WRITE;
  96                 } else {
  97                         /*
  98                          * Check for LDCWX and LDCWS (semaphore instructions).
  99                          * If bits 23 through 25 are all 1's it is one of
 100                          * the above two instructions and is a write.
 101                          *
 102                          * Note: With the limited bits we are looking at,
 103                          * this will also catch PROBEW and PROBEWI. However,
 104                          * these should never get in here because they don't
 105                          * generate exceptions of the type:
 106                          *   Data TLB miss fault/data page fault
 107                          *   Data memory protection trap
 108                          */
 109                         if (bits23_25set(inst) == BITSSET)
 110                                 return VM_WRITE;
 111                 }
 112                 return VM_READ; /* Default */
 113         }
 114         return VM_READ; /* Default */
 115 }
 116
 117 #undef bit22set
 118 #undef bits23_25set
 119 #undef isGraphicsFlushRead
 120 #undef BITSSET
 121
 122
 123 #if 0
 124 /* This is the treewalk to find a vma which is the highest that has
 125  * a start < addr.  We're using find_vma_prev instead right now, but
 126  * we might want to use this at some point in the future.  Probably
 127  * not, but I want it committed to CVS so I don't lose it :-)
 128  */
 129                         while (tree != vm_avl_empty) {
 130                                 if (tree->vm_start > addr) {
 131                                         tree = tree->vm_avl_left;
 132                                 } else {
 133                                         prev = tree;
 134                                         if (prev->vm_next == NULL)
 135                                                 break;
 136                                         if (prev->vm_next->vm_start > addr)
 137                                                 break;
 138                                         tree = tree->vm_avl_right;
 139                                 }
 140                         }
 141 #endif
 142
 143 int fixup_exception(struct pt_regs *regs)
 144 {
 145         const struct exception_table_entry *fix;
 146
 147         fix = search_exception_tables(regs->iaoq[0]);
 148         if (fix) {
 149                 /*
 150                  * Fix up get_user() and put_user().
 151                  * ASM_EXCEPTIONTABLE_ENTRY_EFAULT() sets the least-significant
 152                  * bit in the relative address of the fixup routine to indicate
 153                  * that the register encoded in the "or %r0,%r0,register"
 154                  * opcode should be loaded with -EFAULT to report a userspace
 155                  * access error.
 156                  */
 157                 if (fix->fixup & 1) {
 158                         int fault_error_reg = fix->err_opcode & 0x1f;
 159                         if (!WARN_ON(!fault_error_reg))
 160                                 regs->gr[fault_error_reg] = -EFAULT;
 161                         pr_debug("Unalignment fixup of register %d at %pS\n",
 162                                 fault_error_reg, (void*)regs->iaoq[0]);
 163
 164                         /* zero target register for get_user() */
 165                         if (parisc_acctyp(0, regs->iir) == VM_READ) {
 166                                 int treg = regs->iir & 0x1f;
 167                                 BUG_ON(treg == 0);
 168                                 regs->gr[treg] = 0;
 169                         }
 170                 }
 171
 172                 regs->iaoq[0] = (unsigned long)&fix->fixup + fix->fixup;
 173                 regs->iaoq[0] &= ~3;
 174                 /*
 175                  * NOTE: In some cases the faulting instruction
 176                  * may be in the delay slot of a branch. We
 177                  * don't want to take the branch, so we don't
 178                  * increment iaoq[1], instead we set it to be
 179                  * iaoq[0]+4, and clear the B bit in the PSW
 180                  */
 181                 regs->iaoq[1] = regs->iaoq[0] + 4;
 182                 regs->gr[0] &= ~PSW_B; /* IPSW in gr[0] */
 183
 184                 return 1;
 185         }
 186
 187         return 0;
 188 }
 189
 190 /*
 191  * parisc hardware trap list
 192  *
 193  * Documented in section 3 "Addressing and Access Control" of the
 194  * "PA-RISC 1.1 Architecture and Instruction Set Reference Manual"
 195  * https://parisc.wiki.kernel.org/index.php/File:Pa11_acd.pdf
 196  *
 197  * For implementation see handle_interruption() in traps.c
 198  */
 199 static const char * const trap_description[] = {
 200         [1] =   "High-priority machine check (HPMC)",
 201         [2] =   "Power failure interrupt",
 202         [3] =   "Recovery counter trap",
 203         [5] =   "Low-priority machine check",
 204         [6] =   "Instruction TLB miss fault",
 205         [7] =   "Instruction access rights / protection trap",
 206         [8] =   "Illegal instruction trap",
 207         [9] =   "Break instruction trap",
 208         [10] =  "Privileged operation trap",
 209         [11] =  "Privileged register trap",
 210         [12] =  "Overflow trap",
 211         [13] =  "Conditional trap",
 212         [14] =  "FP Assist Exception trap",
 213         [15] =  "Data TLB miss fault",
 214         [16] =  "Non-access ITLB miss fault",
 215         [17] =  "Non-access DTLB miss fault",
 216         [18] =  "Data memory protection/unaligned access trap",
 217         [19] =  "Data memory break trap",
 218         [20] =  "TLB dirty bit trap",
 219         [21] =  "Page reference trap",
 220         [22] =  "Assist emulation trap",
 221         [25] =  "Taken branch trap",
 222         [26] =  "Data memory access rights trap",
 223         [27] =  "Data memory protection ID trap",
 224         [28] =  "Unaligned data reference trap",
 225 };
 226
 227 const char *trap_name(unsigned long code)
 228 {
 229         const char *t = NULL;
 230
 231         if (code < ARRAY_SIZE(trap_description))
 232                 t = trap_description[code];
 233
 234         return t ? t : "Unknown trap";
 235 }
 236
 237 /*
 238  * Print out info about fatal segfaults, if the show_unhandled_signals
 239  * sysctl is set:
 240  */
 241 static inline void
 242 show_signal_msg(struct pt_regs *regs, unsigned long code,
 243                 unsigned long address, struct task_struct *tsk,
 244                 struct vm_area_struct *vma)
 245 {
 246         if (!unhandled_signal(tsk, SIGSEGV))
 247                 return;
 248
 249         if (!printk_ratelimit())
 250                 return;
 251
 252         pr_warn("\n");
 253         pr_warn("do_page_fault() command='%s' type=%lu address=0x%08lx",
 254             tsk->comm, code, address);
 255         print_vma_addr(KERN_CONT " in ", regs->iaoq[0]);
 256
 257         pr_cont("\ntrap #%lu: %s%c", code, trap_name(code),
 258                 vma ? ',':'\n');
 259
 260         if (vma)
 261                 pr_cont(" vm_start = 0x%08lx, vm_end = 0x%08lx\n",
 262                         vma->vm_start, vma->vm_end);
 263
 264         show_regs(regs);
 265 }
 266
 267 void do_page_fault(struct pt_regs *regs, unsigned long code,
 268                               unsigned long address)
 269 {
 270         struct vm_area_struct *vma, *prev_vma;
 271         struct task_struct *tsk;
 272         struct mm_struct *mm;
 273         unsigned long acc_type;
 274         vm_fault_t fault = 0;
 275         unsigned int flags;
 276         char *msg;
 277
 278         tsk = current;
 279         mm = tsk->mm;
 280         if (!mm) {
 281                 msg = "Page fault: no context";
 282                 goto no_context;
 283         }
 284
 285         flags = FAULT_FLAG_DEFAULT;
 286         if (user_mode(regs))
 287                 flags |= FAULT_FLAG_USER;
 288
 289         acc_type = parisc_acctyp(code, regs->iir);
 290         if (acc_type & VM_WRITE)
 291                 flags |= FAULT_FLAG_WRITE;
 292         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
 293 retry:
 294         mmap_read_lock(mm);
 295         vma = find_vma_prev(mm, address, &prev_vma);
 296         if (!vma || address < vma->vm_start) {
 297                 if (!prev_vma || !(prev_vma->vm_flags & VM_GROWSUP))
 298                         goto bad_area;
 299                 vma = expand_stack(mm, address);
 300                 if (!vma)
 301                         goto bad_area_nosemaphore;
 302         }
 303
 304 /*
 305  * Ok, we have a good vm_area for this memory access. We still need to
 306  * check the access permissions.
 307  */
 308
 309         if ((vma->vm_flags & acc_type) != acc_type)
 310                 goto bad_area;
 311
 312         /*
 313          * If for any reason at all we couldn't handle the fault, make
 314          * sure we exit gracefully rather than endlessly redo the
 315          * fault.
 316          */
 317
 318         fault = handle_mm_fault(vma, address, flags, regs);
 319
 320         if (fault_signal_pending(fault, regs)) {
 321                 if (!user_mode(regs)) {
 322                         msg = "Page fault: fault signal on kernel memory";
 323                         goto no_context;
 324                 }
 325                 return;
 326         }
 327
 328         /* The fault is fully completed (including releasing mmap lock) */
 329         if (fault & VM_FAULT_COMPLETED)
 330                 return;
 331
 332         if (unlikely(fault & VM_FAULT_ERROR)) {
 333                 /*
 334                  * We hit a shared mapping outside of the file, or some
 335                  * other thing happened to us that made us unable to
 336                  * handle the page fault gracefully.
 337                  */
 338                 if (fault & VM_FAULT_OOM)
 339                         goto out_of_memory;
 340                 else if (fault & VM_FAULT_SIGSEGV)
 341                         goto bad_area;
 342                 else if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
 343                                   VM_FAULT_HWPOISON_LARGE))
 344                         goto bad_area;
 345                 BUG();
 346         }
 347         if (fault & VM_FAULT_RETRY) {
 348                 /*
 349                  * No need to mmap_read_unlock(mm) as we would
 350                  * have already released it in __lock_page_or_retry
 351                  * in mm/filemap.c.
 352                  */
 353                 flags |= FAULT_FLAG_TRIED;
 354                 goto retry;
 355         }
 356         mmap_read_unlock(mm);
 357         return;
 358
 359 /*
 360  * Something tried to access memory that isn't in our memory map..
 361  */
 362 bad_area:
 363         mmap_read_unlock(mm);
 364
 365 bad_area_nosemaphore:
 366         if (user_mode(regs)) {
 367                 int signo, si_code;
 368
 369                 switch (code) {
 370                 case 15:        /* Data TLB miss fault/Data page fault */
 371                         /* send SIGSEGV when outside of vma */
 372                         if (!vma ||
 373                             address < vma->vm_start || address >= vma->vm_end) {
 374                                 signo = SIGSEGV;
 375                                 si_code = SEGV_MAPERR;
 376                                 break;
 377                         }
 378
 379                         /* send SIGSEGV for wrong permissions */
 380                         if ((vma->vm_flags & acc_type) != acc_type) {
 381                                 signo = SIGSEGV;
 382                                 si_code = SEGV_ACCERR;
 383                                 break;
 384                         }
 385
 386                         /* probably address is outside of mapped file */
 387                         fallthrough;
 388                 case 17:        /* NA data TLB miss / page fault */
 389                 case 18:        /* Unaligned access - PCXS only */
 390                         signo = SIGBUS;
 391                         si_code = (code == 18) ? BUS_ADRALN : BUS_ADRERR;
 392                         break;
 393                 case 16:        /* Non-access instruction TLB miss fault */
 394                 case 26:        /* PCXL: Data memory access rights trap */
 395                 default:
 396                         signo = SIGSEGV;
 397                         si_code = (code == 26) ? SEGV_ACCERR : SEGV_MAPERR;
 398                         break;
 399                 }
 400 #ifdef CONFIG_MEMORY_FAILURE
 401                 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
 402                         unsigned int lsb = 0;
 403                         printk(KERN_ERR
 404         "MCE: Killing %s:%d due to hardware memory corruption fault at %08lx\n",
 405                         tsk->comm, tsk->pid, address);
 406                         /*
 407                          * Either small page or large page may be poisoned.
 408                          * In other words, VM_FAULT_HWPOISON_LARGE and
 409                          * VM_FAULT_HWPOISON are mutually exclusive.
 410                          */
 411                         if (fault & VM_FAULT_HWPOISON_LARGE)
 412                                 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
 413                         else if (fault & VM_FAULT_HWPOISON)
 414                                 lsb = PAGE_SHIFT;
 415
 416                         force_sig_mceerr(BUS_MCEERR_AR, (void __user *) address,
 417                                          lsb);
 418                         return;
 419                 }
 420 #endif
 421                 show_signal_msg(regs, code, address, tsk, vma);
 422
 423                 force_sig_fault(signo, si_code, (void __user *) address);
 424                 return;
 425         }
 426         msg = "Page fault: bad address";
 427
 428 no_context:
 429
 430         if (!user_mode(regs) && fixup_exception(regs)) {
 431                 return;
 432         }
 433
 434         parisc_terminate(msg, regs, code, address);
 435
 436 out_of_memory:
 437         mmap_read_unlock(mm);
 438         if (!user_mode(regs)) {
 439                 msg = "Page fault: out of memory";
 440                 goto no_context;
 441         }
 442         pagefault_out_of_memory();
 443 }
 444
 445 /* Handle non-access data TLB miss faults.
 446  *
 447  * For probe instructions, accesses to userspace are considered allowed
 448  * if they lie in a valid VMA and the access type matches. We are not
 449  * allowed to handle MM faults here so there may be situations where an
 450  * actual access would fail even though a probe was successful.
 451  */
 452 int
 453 handle_nadtlb_fault(struct pt_regs *regs)
 454 {
 455         unsigned long insn = regs->iir;
 456         int breg, treg, xreg, val = 0;
 457         struct vm_area_struct *vma;
 458         struct task_struct *tsk;
 459         struct mm_struct *mm;
 460         unsigned long address;
 461         unsigned long acc_type;
 462
 463         switch (insn & 0x380) {
 464         case 0x280:
 465                 /* FDC instruction */
 466                 fallthrough;
 467         case 0x380:
 468                 /* PDC and FIC instructions */
 469                 if (DEBUG_NATLB && printk_ratelimit()) {
 470                         pr_warn("WARNING: nullifying cache flush/purge instruction\n");
 471                         show_regs(regs);
 472                 }
 473                 if (insn & 0x20) {
 474                         /* Base modification */
 475                         breg = (insn >> 21) & 0x1f;
 476                         xreg = (insn >> 16) & 0x1f;
 477                         if (breg && xreg)
 478                                 regs->gr[breg] += regs->gr[xreg];
 479                 }
 480                 regs->gr[0] |= PSW_N;
 481                 return 1;
 482
 483         case 0x180:
 484                 /* PROBE instruction */
 485                 treg = insn & 0x1f;
 486                 if (regs->isr) {
 487                         tsk = current;
 488                         mm = tsk->mm;
 489                         if (mm) {
 490                                 /* Search for VMA */
 491                                 address = regs->ior;
 492                                 mmap_read_lock(mm);
 493                                 vma = vma_lookup(mm, address);
 494                                 mmap_read_unlock(mm);
 495
 496                                 /*
 497                                  * Check if access to the VMA is okay.
 498                                  * We don't allow for stack expansion.
 499                                  */
 500                                 acc_type = (insn & 0x40) ? VM_WRITE : VM_READ;
 501                                 if (vma
 502                                     && (vma->vm_flags & acc_type) == acc_type)
 503                                         val = 1;
 504                         }
 505                 }
 506                 if (treg)
 507                         regs->gr[treg] = val;
 508                 regs->gr[0] |= PSW_N;
 509                 return 1;
 510
 511         case 0x300:
 512                 /* LPA instruction */
 513                 if (insn & 0x20) {
 514                         /* Base modification */
 515                         breg = (insn >> 21) & 0x1f;
 516                         xreg = (insn >> 16) & 0x1f;
 517                         if (breg && xreg)
 518                                 regs->gr[breg] += regs->gr[xreg];
 519                 }
 520                 treg = insn & 0x1f;
 521                 if (treg)
 522                         regs->gr[treg] = 0;
 523                 regs->gr[0] |= PSW_N;
 524                 return 1;
 525
 526         default:
 527                 break;
 528         }
 529
 530         return 0;
 531 }