arch/openrisc/mm/fault.c

   1 /*
   2  * OpenRISC fault.c
   3  *
   4  * Linux architectural port borrowing liberally from similar works of
   5  * others.  All original copyrights apply as per the original source
   6  * declaration.
   7  *
   8  * Modifications for the OpenRISC architecture:
   9  * Copyright (C) 2003 Matjaz Breskvar <phoenix@bsemi.com>
  10  * Copyright (C) 2010-2011 Jonas Bonn <jonas@southpole.se>
  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/mm.h>
  19 #include <linux/interrupt.h>
  20 #include <linux/module.h>
  21 #include <linux/sched.h>
  22
  23 #include <asm/uaccess.h>
  24 #include <asm/siginfo.h>
  25 #include <asm/signal.h>
  26
  27 #define NUM_TLB_ENTRIES 64
  28 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
  29
  30 unsigned long pte_misses;       /* updated by do_page_fault() */
  31 unsigned long pte_errors;       /* updated by do_page_fault() */
  32
  33 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
  34  *            - also look into include/asm-or32/mmu_context.h
  35  */
  36 volatile pgd_t *current_pgd;
  37
  38 extern void die(char *, struct pt_regs *, long);
  39
  40 /*
  41  * This routine handles page faults.  It determines the address,
  42  * and the problem, and then passes it off to one of the appropriate
  43  * routines.
  44  *
  45  * If this routine detects a bad access, it returns 1, otherwise it
  46  * returns 0.
  47  */
  48
  49 asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long address,
  50                               unsigned long vector, int write_acc)
  51 {
  52         struct task_struct *tsk;
  53         struct mm_struct *mm;
  54         struct vm_area_struct *vma;
  55         siginfo_t info;
  56         int fault;
  57         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  58
  59         tsk = current;
  60
  61         /*
  62          * We fault-in kernel-space virtual memory on-demand. The
  63          * 'reference' page table is init_mm.pgd.
  64          *
  65          * NOTE! We MUST NOT take any locks for this case. We may
  66          * be in an interrupt or a critical region, and should
  67          * only copy the information from the master page table,
  68          * nothing more.
  69          *
  70          * NOTE2: This is done so that, when updating the vmalloc
  71          * mappings we don't have to walk all processes pgdirs and
  72          * add the high mappings all at once. Instead we do it as they
  73          * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
  74          * bit set so sometimes the TLB can use a lingering entry.
  75          *
  76          * This verifies that the fault happens in kernel space
  77          * and that the fault was not a protection error.
  78          */
  79
  80         if (address >= VMALLOC_START &&
  81             (vector != 0x300 && vector != 0x400) &&
  82             !user_mode(regs))
  83                 goto vmalloc_fault;
  84
  85         /* If exceptions were enabled, we can reenable them here */
  86         if (user_mode(regs)) {
  87                 /* Exception was in userspace: reenable interrupts */
  88                 local_irq_enable();
  89         } else {
  90                 /* If exception was in a syscall, then IRQ's may have
  91                  * been enabled or disabled.  If they were enabled,
  92                  * reenable them.
  93                  */
  94                 if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
  95                         local_irq_enable();
  96         }
  97
  98         mm = tsk->mm;
  99         info.si_code = SEGV_MAPERR;
 100
 101         /*
 102          * If we're in an interrupt or have no user
 103          * context, we must not take the fault..
 104          */
 105
 106         if (in_interrupt() || !mm)
 107                 goto no_context;
 108
 109 retry:
 110         down_read(&mm->mmap_sem);
 111         vma = find_vma(mm, address);
 112
 113         if (!vma)
 114                 goto bad_area;
 115
 116         if (vma->vm_start <= address)
 117                 goto good_area;
 118
 119         if (!(vma->vm_flags & VM_GROWSDOWN))
 120                 goto bad_area;
 121
 122         if (user_mode(regs)) {
 123                 /*
 124                  * accessing the stack below usp is always a bug.
 125                  * we get page-aligned addresses so we can only check
 126                  * if we're within a page from usp, but that might be
 127                  * enough to catch brutal errors at least.
 128                  */
 129                 if (address + PAGE_SIZE < regs->sp)
 130                         goto bad_area;
 131         }
 132         if (expand_stack(vma, address))
 133                 goto bad_area;
 134
 135         /*
 136          * Ok, we have a good vm_area for this memory access, so
 137          * we can handle it..
 138          */
 139
 140 good_area:
 141         info.si_code = SEGV_ACCERR;
 142
 143         /* first do some preliminary protection checks */
 144
 145         if (write_acc) {
 146                 if (!(vma->vm_flags & VM_WRITE))
 147                         goto bad_area;
 148                 flags |= FAULT_FLAG_WRITE;
 149         } else {
 150                 /* not present */
 151                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
 152                         goto bad_area;
 153         }
 154
 155         /* are we trying to execute nonexecutable area */
 156         if ((vector == 0x400) && !(vma->vm_page_prot.pgprot & _PAGE_EXEC))
 157                 goto bad_area;
 158
 159         /*
 160          * If for any reason at all we couldn't handle the fault,
 161          * make sure we exit gracefully rather than endlessly redo
 162          * the fault.
 163          */
 164
 165         fault = handle_mm_fault(mm, vma, address, flags);
 166
 167         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 168                 return;
 169
 170         if (unlikely(fault & VM_FAULT_ERROR)) {
 171                 if (fault & VM_FAULT_OOM)
 172                         goto out_of_memory;
 173                 else if (fault & VM_FAULT_SIGBUS)
 174                         goto do_sigbus;
 175                 BUG();
 176         }
 177
 178         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 179                 /*RGD modeled on Cris */
 180                 if (fault & VM_FAULT_MAJOR)
 181                         tsk->maj_flt++;
 182                 else
 183                         tsk->min_flt++;
 184                 if (fault & VM_FAULT_RETRY) {
 185                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 186
 187                          /* No need to up_read(&mm->mmap_sem) as we would
 188                          * have already released it in __lock_page_or_retry
 189                          * in mm/filemap.c.
 190                          */
 191
 192                         goto retry;
 193                 }
 194         }
 195
 196         up_read(&mm->mmap_sem);
 197         return;
 198
 199         /*
 200          * Something tried to access memory that isn't in our memory map..
 201          * Fix it, but check if it's kernel or user first..
 202          */
 203
 204 bad_area:
 205         up_read(&mm->mmap_sem);
 206
 207 bad_area_nosemaphore:
 208
 209         /* User mode accesses just cause a SIGSEGV */
 210
 211         if (user_mode(regs)) {
 212                 info.si_signo = SIGSEGV;
 213                 info.si_errno = 0;
 214                 /* info.si_code has been set above */
 215                 info.si_addr = (void *)address;
 216                 force_sig_info(SIGSEGV, &info, tsk);
 217                 return;
 218         }
 219
 220 no_context:
 221
 222         /* Are we prepared to handle this kernel fault?
 223          *
 224          * (The kernel has valid exception-points in the source
 225          *  when it acesses user-memory. When it fails in one
 226          *  of those points, we find it in a table and do a jump
 227          *  to some fixup code that loads an appropriate error
 228          *  code)
 229          */
 230
 231         {
 232                 const struct exception_table_entry *entry;
 233
 234                 __asm__ __volatile__("l.nop 42");
 235
 236                 if ((entry = search_exception_tables(regs->pc)) != NULL) {
 237                         /* Adjust the instruction pointer in the stackframe */
 238                         regs->pc = entry->fixup;
 239                         return;
 240                 }
 241         }
 242
 243         /*
 244          * Oops. The kernel tried to access some bad page. We'll have to
 245          * terminate things with extreme prejudice.
 246          */
 247
 248         if ((unsigned long)(address) < PAGE_SIZE)
 249                 printk(KERN_ALERT
 250                        "Unable to handle kernel NULL pointer dereference");
 251         else
 252                 printk(KERN_ALERT "Unable to handle kernel access");
 253         printk(" at virtual address 0x%08lx\n", address);
 254
 255         die("Oops", regs, write_acc);
 256
 257         do_exit(SIGKILL);
 258
 259         /*
 260          * We ran out of memory, or some other thing happened to us that made
 261          * us unable to handle the page fault gracefully.
 262          */
 263
 264 out_of_memory:
 265         __asm__ __volatile__("l.nop 42");
 266         __asm__ __volatile__("l.nop 1");
 267
 268         up_read(&mm->mmap_sem);
 269         printk("VM: killing process %s\n", tsk->comm);
 270         if (user_mode(regs))
 271                 do_exit(SIGKILL);
 272         goto no_context;
 273
 274 do_sigbus:
 275         up_read(&mm->mmap_sem);
 276
 277         /*
 278          * Send a sigbus, regardless of whether we were in kernel
 279          * or user mode.
 280          */
 281         info.si_signo = SIGBUS;
 282         info.si_errno = 0;
 283         info.si_code = BUS_ADRERR;
 284         info.si_addr = (void *)address;
 285         force_sig_info(SIGBUS, &info, tsk);
 286
 287         /* Kernel mode? Handle exceptions or die */
 288         if (!user_mode(regs))
 289                 goto no_context;
 290         return;
 291
 292 vmalloc_fault:
 293         {
 294                 /*
 295                  * Synchronize this task's top level page-table
 296                  * with the 'reference' page table.
 297                  *
 298                  * Use current_pgd instead of tsk->active_mm->pgd
 299                  * since the latter might be unavailable if this
 300                  * code is executed in a misfortunately run irq
 301                  * (like inside schedule() between switch_mm and
 302                  *  switch_to...).
 303                  */
 304
 305                 int offset = pgd_index(address);
 306                 pgd_t *pgd, *pgd_k;
 307                 pud_t *pud, *pud_k;
 308                 pmd_t *pmd, *pmd_k;
 309                 pte_t *pte_k;
 310
 311 /*
 312                 phx_warn("do_page_fault(): vmalloc_fault will not work, "
 313                          "since current_pgd assign a proper value somewhere\n"
 314                          "anyhow we don't need this at the moment\n");
 315
 316                 phx_mmu("vmalloc_fault");
 317 */
 318                 pgd = (pgd_t *)current_pgd + offset;
 319                 pgd_k = init_mm.pgd + offset;
 320
 321                 /* Since we're two-level, we don't need to do both
 322                  * set_pgd and set_pmd (they do the same thing). If
 323                  * we go three-level at some point, do the right thing
 324                  * with pgd_present and set_pgd here.
 325                  *
 326                  * Also, since the vmalloc area is global, we don't
 327                  * need to copy individual PTE's, it is enough to
 328                  * copy the pgd pointer into the pte page of the
 329                  * root task. If that is there, we'll find our pte if
 330                  * it exists.
 331                  */
 332
 333                 pud = pud_offset(pgd, address);
 334                 pud_k = pud_offset(pgd_k, address);
 335                 if (!pud_present(*pud_k))
 336                         goto no_context;
 337
 338                 pmd = pmd_offset(pud, address);
 339                 pmd_k = pmd_offset(pud_k, address);
 340
 341                 if (!pmd_present(*pmd_k))
 342                         goto bad_area_nosemaphore;
 343
 344                 set_pmd(pmd, *pmd_k);
 345
 346                 /* Make sure the actual PTE exists as well to
 347                  * catch kernel vmalloc-area accesses to non-mapped
 348                  * addresses. If we don't do this, this will just
 349                  * silently loop forever.
 350                  */
 351
 352                 pte_k = pte_offset_kernel(pmd_k, address);
 353                 if (!pte_present(*pte_k))
 354                         goto no_context;
 355
 356                 return;
 357         }
 358 }