arch/ppc/mm/fault.c

   1 /*
   2  *  arch/ppc/mm/fault.c
   3  *
   4  *  PowerPC version
   5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   6  *
   7  *  Derived from "arch/i386/mm/fault.c"
   8  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   9  *
  10  *  Modified by Cort Dougan and Paul Mackerras.
  11  *
  12  *  This program is free software; you can redistribute it and/or
  13  *  modify it under the terms of the GNU General Public License
  14  *  as published by the Free Software Foundation; either version
  15  *  2 of the License, or (at your option) any later version.
  16  */
  17
  18 #include <linux/config.h>
  19 #include <linux/signal.h>
  20 #include <linux/sched.h>
  21 #include <linux/kernel.h>
  22 #include <linux/errno.h>
  23 #include <linux/string.h>
  24 #include <linux/types.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/mman.h>
  27 #include <linux/mm.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/highmem.h>
  30 #include <linux/module.h>
  31
  32 #include <asm/page.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/mmu.h>
  35 #include <asm/mmu_context.h>
  36 #include <asm/system.h>
  37 #include <asm/uaccess.h>
  38 #include <asm/tlbflush.h>
  39
  40 #if defined(CONFIG_XMON) || defined(CONFIG_KGDB)
  41 extern void (*debugger)(struct pt_regs *);
  42 extern void (*debugger_fault_handler)(struct pt_regs *);
  43 extern int (*debugger_dabr_match)(struct pt_regs *);
  44 int debugger_kernel_faults = 1;
  45 #endif
  46
  47 unsigned long htab_reloads;     /* updated by hashtable.S:hash_page() */
  48 unsigned long htab_evicts;      /* updated by hashtable.S:hash_page() */
  49 unsigned long htab_preloads;    /* updated by hashtable.S:add_hash_page() */
  50 unsigned long pte_misses;       /* updated by do_page_fault() */
  51 unsigned long pte_errors;       /* updated by do_page_fault() */
  52 unsigned int probingmem;
  53
  54 /*
  55  * Check whether the instruction at regs->nip is a store using
  56  * an update addressing form which will update r1.
  57  */
  58 static int store_updates_sp(struct pt_regs *regs)
  59 {
  60         unsigned int inst;
  61
  62         if (get_user(inst, (unsigned int __user *)regs->nip))
  63                 return 0;
  64         /* check for 1 in the rA field */
  65         if (((inst >> 16) & 0x1f) != 1)
  66                 return 0;
  67         /* check major opcode */
  68         switch (inst >> 26) {
  69         case 37:        /* stwu */
  70         case 39:        /* stbu */
  71         case 45:        /* sthu */
  72         case 53:        /* stfsu */
  73         case 55:        /* stfdu */
  74                 return 1;
  75         case 31:
  76                 /* check minor opcode */
  77                 switch ((inst >> 1) & 0x3ff) {
  78                 case 183:       /* stwux */
  79                 case 247:       /* stbux */
  80                 case 439:       /* sthux */
  81                 case 695:       /* stfsux */
  82                 case 759:       /* stfdux */
  83                         return 1;
  84                 }
  85         }
  86         return 0;
  87 }
  88
  89 /*
  90  * For 600- and 800-family processors, the error_code parameter is DSISR
  91  * for a data fault, SRR1 for an instruction fault. For 400-family processors
  92  * the error_code parameter is ESR for a data fault, 0 for an instruction
  93  * fault.
  94  */
  95 int do_page_fault(struct pt_regs *regs, unsigned long address,
  96                   unsigned long error_code)
  97 {
  98         struct vm_area_struct * vma;
  99         struct mm_struct *mm = current->mm;
 100         siginfo_t info;
 101         int code = SEGV_MAPERR;
 102 #if defined(CONFIG_4xx) || defined (CONFIG_BOOKE)
 103         int is_write = error_code & ESR_DST;
 104 #else
 105         int is_write = 0;
 106
 107         /*
 108          * Fortunately the bit assignments in SRR1 for an instruction
 109          * fault and DSISR for a data fault are mostly the same for the
 110          * bits we are interested in.  But there are some bits which
 111          * indicate errors in DSISR but can validly be set in SRR1.
 112          */
 113         if (TRAP(regs) == 0x400)
 114                 error_code &= 0x48200000;
 115         else
 116                 is_write = error_code & 0x02000000;
 117 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 118
 119 #if defined(CONFIG_XMON) || defined(CONFIG_KGDB)
 120         if (debugger_fault_handler && TRAP(regs) == 0x300) {
 121                 debugger_fault_handler(regs);
 122                 return 0;
 123         }
 124 #if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
 125         if (error_code & 0x00400000) {
 126                 /* DABR match */
 127                 if (debugger_dabr_match(regs))
 128                         return 0;
 129         }
 130 #endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
 131 #endif /* CONFIG_XMON || CONFIG_KGDB */
 132
 133         if (in_atomic() || mm == NULL)
 134                 return SIGSEGV;
 135
 136         down_read(&mm->mmap_sem);
 137         vma = find_vma(mm, address);
 138         if (!vma)
 139                 goto bad_area;
 140         if (vma->vm_start <= address)
 141                 goto good_area;
 142         if (!(vma->vm_flags & VM_GROWSDOWN))
 143                 goto bad_area;
 144         if (!is_write)
 145                 goto bad_area;
 146
 147         /*
 148          * N.B. The rs6000/xcoff ABI allows programs to access up to
 149          * a few hundred bytes below the stack pointer.
 150          * The kernel signal delivery code writes up to about 1.5kB
 151          * below the stack pointer (r1) before decrementing it.
 152          * The exec code can write slightly over 640kB to the stack
 153          * before setting the user r1.  Thus we allow the stack to
 154          * expand to 1MB without further checks.
 155          */
 156         if (address + 0x100000 < vma->vm_end) {
 157                 /* get user regs even if this fault is in kernel mode */
 158                 struct pt_regs *uregs = current->thread.regs;
 159                 if (uregs == NULL)
 160                         goto bad_area;
 161
 162                 /*
 163                  * A user-mode access to an address a long way below
 164                  * the stack pointer is only valid if the instruction
 165                  * is one which would update the stack pointer to the
 166                  * address accessed if the instruction completed,
 167                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 168                  * (or the byte, halfword, float or double forms).
 169                  *
 170                  * If we don't check this then any write to the area
 171                  * between the last mapped region and the stack will
 172                  * expand the stack rather than segfaulting.
 173                  */
 174                 if (address + 2048 < uregs->gpr[1]
 175                     && (!user_mode(regs) || !store_updates_sp(regs)))
 176                         goto bad_area;
 177         }
 178         if (expand_stack(vma, address))
 179                 goto bad_area;
 180
 181 good_area:
 182         code = SEGV_ACCERR;
 183 #if defined(CONFIG_6xx)
 184         if (error_code & 0x95700000)
 185                 /* an error such as lwarx to I/O controller space,
 186                    address matching DABR, eciwx, etc. */
 187                 goto bad_area;
 188 #endif /* CONFIG_6xx */
 189 #if defined(CONFIG_8xx)
 190         /* The MPC8xx seems to always set 0x80000000, which is
 191          * "undefined".  Of those that can be set, this is the only
 192          * one which seems bad.
 193          */
 194         if (error_code & 0x10000000)
 195                 /* Guarded storage error. */
 196                 goto bad_area;
 197 #endif /* CONFIG_8xx */
 198
 199         /* a write */
 200         if (is_write) {
 201                 if (!(vma->vm_flags & VM_WRITE))
 202                         goto bad_area;
 203 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
 204         /* an exec  - 4xx/Book-E allows for per-page execute permission */
 205         } else if (TRAP(regs) == 0x400) {
 206                 pte_t *ptep;
 207
 208 #if 0
 209                 /* It would be nice to actually enforce the VM execute
 210                    permission on CPUs which can do so, but far too
 211                    much stuff in userspace doesn't get the permissions
 212                    right, so we let any page be executed for now. */
 213                 if (! (vma->vm_flags & VM_EXEC))
 214                         goto bad_area;
 215 #endif
 216
 217                 /* Since 4xx/Book-E supports per-page execute permission,
 218                  * we lazily flush dcache to icache. */
 219                 ptep = NULL;
 220                 if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
 221                         struct page *page = pte_page(*ptep);
 222
 223                         if (! test_bit(PG_arch_1, &page->flags)) {
 224                                 flush_dcache_icache_page(page);
 225                                 set_bit(PG_arch_1, &page->flags);
 226                         }
 227                         pte_update(ptep, 0, _PAGE_HWEXEC);
 228                         _tlbie(address);
 229                         pte_unmap(ptep);
 230                         up_read(&mm->mmap_sem);
 231                         return 0;
 232                 }
 233                 if (ptep != NULL)
 234                         pte_unmap(ptep);
 235 #endif
 236         /* a read */
 237         } else {
 238                 /* protection fault */
 239                 if (error_code & 0x08000000)
 240                         goto bad_area;
 241                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 242                         goto bad_area;
 243         }
 244
 245         /*
 246          * If for any reason at all we couldn't handle the fault,
 247          * make sure we exit gracefully rather than endlessly redo
 248          * the fault.
 249          */
 250  survive:
 251         switch (handle_mm_fault(mm, vma, address, is_write)) {
 252         case VM_FAULT_MINOR:
 253                 current->min_flt++;
 254                 break;
 255         case VM_FAULT_MAJOR:
 256                 current->maj_flt++;
 257                 break;
 258         case VM_FAULT_SIGBUS:
 259                 goto do_sigbus;
 260         case VM_FAULT_OOM:
 261                 goto out_of_memory;
 262         default:
 263                 BUG();
 264         }
 265
 266         up_read(&mm->mmap_sem);
 267         /*
 268          * keep track of tlb+htab misses that are good addrs but
 269          * just need pte's created via handle_mm_fault()
 270          * -- Cort
 271          */
 272         pte_misses++;
 273         return 0;
 274
 275 bad_area:
 276         up_read(&mm->mmap_sem);
 277         pte_errors++;
 278
 279         /* User mode accesses cause a SIGSEGV */
 280         if (user_mode(regs)) {
 281                 _exception(SIGSEGV, regs, code, address);
 282                 return 0;
 283         }
 284
 285         return SIGSEGV;
 286
 287 /*
 288  * We ran out of memory, or some other thing happened to us that made
 289  * us unable to handle the page fault gracefully.
 290  */
 291 out_of_memory:
 292         up_read(&mm->mmap_sem);
 293         if (current->pid == 1) {
 294                 yield();
 295                 down_read(&mm->mmap_sem);
 296                 goto survive;
 297         }
 298         printk("VM: killing process %s\n", current->comm);
 299         if (user_mode(regs))
 300                 do_exit(SIGKILL);
 301         return SIGKILL;
 302
 303 do_sigbus:
 304         up_read(&mm->mmap_sem);
 305         info.si_signo = SIGBUS;
 306         info.si_errno = 0;
 307         info.si_code = BUS_ADRERR;
 308         info.si_addr = (void __user *)address;
 309         force_sig_info (SIGBUS, &info, current);
 310         if (!user_mode(regs))
 311                 return SIGBUS;
 312         return 0;
 313 }
 314
 315 /*
 316  * bad_page_fault is called when we have a bad access from the kernel.
 317  * It is called from the DSI and ISI handlers in head.S and from some
 318  * of the procedures in traps.c.
 319  */
 320 void
 321 bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 322 {
 323         const struct exception_table_entry *entry;
 324
 325         /* Are we prepared to handle this fault?  */
 326         if ((entry = search_exception_tables(regs->nip)) != NULL) {
 327                 regs->nip = entry->fixup;
 328                 return;
 329         }
 330
 331         /* kernel has accessed a bad area */
 332 #if defined(CONFIG_XMON) || defined(CONFIG_KGDB)
 333         if (debugger_kernel_faults)
 334                 debugger(regs);
 335 #endif
 336         die("kernel access of bad area", regs, sig);
 337 }
 338
 339 #ifdef CONFIG_8xx
 340
 341 /* The pgtable.h claims some functions generically exist, but I
 342  * can't find them......
 343  */
 344 pte_t *va_to_pte(unsigned long address)
 345 {
 346         pgd_t *dir;
 347         pmd_t *pmd;
 348         pte_t *pte;
 349
 350         if (address < TASK_SIZE)
 351                 return NULL;
 352
 353         dir = pgd_offset(&init_mm, address);
 354         if (dir) {
 355                 pmd = pmd_offset(dir, address & PAGE_MASK);
 356                 if (pmd && pmd_present(*pmd)) {
 357                         pte = pte_offset_kernel(pmd, address & PAGE_MASK);
 358                         if (pte && pte_present(*pte))
 359                                 return(pte);
 360                 }
 361         }
 362         return NULL;
 363 }
 364
 365 unsigned long va_to_phys(unsigned long address)
 366 {
 367         pte_t *pte;
 368
 369         pte = va_to_pte(address);
 370         if (pte)
 371                 return(((unsigned long)(pte_val(*pte)) & PAGE_MASK) | (address & ~(PAGE_MASK)));
 372         return (0);
 373 }
 374
 375 void
 376 print_8xx_pte(struct mm_struct *mm, unsigned long addr)
 377 {
 378         pgd_t * pgd;
 379         pmd_t * pmd;
 380         pte_t * pte;
 381
 382         printk(" pte @ 0x%8lx: ", addr);
 383         pgd = pgd_offset(mm, addr & PAGE_MASK);
 384         if (pgd) {
 385                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
 386                 if (pmd && pmd_present(*pmd)) {
 387                         pte = pte_offset_kernel(pmd, addr & PAGE_MASK);
 388                         if (pte) {
 389                                 printk(" (0x%08lx)->(0x%08lx)->0x%08lx\n",
 390                                         (long)pgd, (long)pte, (long)pte_val(*pte));
 391 #define pp ((long)pte_val(*pte))
 392                                 printk(" RPN: %05lx PP: %lx SPS: %lx SH: %lx "
 393                                        "CI: %lx v: %lx\n",
 394                                        pp>>12,    /* rpn */
 395                                        (pp>>10)&3, /* pp */
 396                                        (pp>>3)&1, /* small */
 397                                        (pp>>2)&1, /* shared */
 398                                        (pp>>1)&1, /* cache inhibit */
 399                                        pp&1       /* valid */
 400                                        );
 401 #undef pp
 402                         }
 403                         else {
 404                                 printk("no pte\n");
 405                         }
 406                 }
 407                 else {
 408                         printk("no pmd\n");
 409                 }
 410         }
 411         else {
 412                 printk("no pgd\n");
 413         }
 414 }
 415
 416 int
 417 get_8xx_pte(struct mm_struct *mm, unsigned long addr)
 418 {
 419         pgd_t * pgd;
 420         pmd_t * pmd;
 421         pte_t * pte;
 422         int     retval = 0;
 423
 424         pgd = pgd_offset(mm, addr & PAGE_MASK);
 425         if (pgd) {
 426                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
 427                 if (pmd && pmd_present(*pmd)) {
 428                         pte = pte_offset_kernel(pmd, addr & PAGE_MASK);
 429                         if (pte) {
 430                                 retval = (int)pte_val(*pte);
 431                         }
 432                 }
 433         }
 434         return(retval);
 435 }
 436 #endif /* CONFIG_8xx */