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
  58         tsk = current;
  59
  60         /*
  61          * We fault-in kernel-space virtual memory on-demand. The
  62          * 'reference' page table is init_mm.pgd.
  63          *
  64          * NOTE! We MUST NOT take any locks for this case. We may
  65          * be in an interrupt or a critical region, and should
  66          * only copy the information from the master page table,
  67          * nothing more.
  68          *
  69          * NOTE2: This is done so that, when updating the vmalloc
  70          * mappings we don't have to walk all processes pgdirs and
  71          * add the high mappings all at once. Instead we do it as they
  72          * are used. However vmalloc'ed page entries have the PAGE_GLOBAL
  73          * bit set so sometimes the TLB can use a lingering entry.
  74          *
  75          * This verifies that the fault happens in kernel space
  76          * and that the fault was not a protection error.
  77          */
  78
  79         if (address >= VMALLOC_START &&
  80             (vector != 0x300 && vector != 0x400) &&
  81             !user_mode(regs))
  82                 goto vmalloc_fault;
  83
  84         /* If exceptions were enabled, we can reenable them here */
  85         if (user_mode(regs)) {
  86                 /* Exception was in userspace: reenable interrupts */
  87                 local_irq_enable();
  88         } else {
  89                 /* If exception was in a syscall, then IRQ's may have
  90                  * been enabled or disabled.  If they were enabled,
  91                  * reenable them.
  92                  */
  93                 if (regs->sr && (SPR_SR_IEE | SPR_SR_TEE))
  94                         local_irq_enable();
  95         }
  96
  97         mm = tsk->mm;
  98         info.si_code = SEGV_MAPERR;
  99
 100         /*
 101          * If we're in an interrupt or have no user
 102          * context, we must not take the fault..
 103          */
 104
 105         if (in_interrupt() || !mm)
 106                 goto no_context;
 107
 108         down_read(&mm->mmap_sem);
 109         vma = find_vma(mm, address);
 110
 111         if (!vma)
 112                 goto bad_area;
 113
 114         if (vma->vm_start <= address)
 115                 goto good_area;
 116
 117         if (!(vma->vm_flags & VM_GROWSDOWN))
 118                 goto bad_area;
 119
 120         if (user_mode(regs)) {
 121                 /*
 122                  * accessing the stack below usp is always a bug.
 123                  * we get page-aligned addresses so we can only check
 124                  * if we're within a page from usp, but that might be
 125                  * enough to catch brutal errors at least.
 126                  */
 127                 if (address + PAGE_SIZE < regs->sp)
 128                         goto bad_area;
 129         }
 130         if (expand_stack(vma, address))
 131                 goto bad_area;
 132
 133         /*
 134          * Ok, we have a good vm_area for this memory access, so
 135          * we can handle it..
 136          */
 137
 138 good_area:
 139         info.si_code = SEGV_ACCERR;
 140
 141         /* first do some preliminary protection checks */
 142
 143         if (write_acc) {
 144                 if (!(vma->vm_flags & VM_WRITE))
 145                         goto bad_area;
 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(mm, vma, address, write_acc);
 163         if (unlikely(fault & VM_FAULT_ERROR)) {
 164                 if (fault & VM_FAULT_OOM)
 165                         goto out_of_memory;
 166                 else if (fault & VM_FAULT_SIGBUS)
 167                         goto do_sigbus;
 168                 BUG();
 169         }
 170         /*RGD modeled on Cris */
 171         if (fault & VM_FAULT_MAJOR)
 172                 tsk->maj_flt++;
 173         else
 174                 tsk->min_flt++;
 175
 176         up_read(&mm->mmap_sem);
 177         return;
 178
 179         /*
 180          * Something tried to access memory that isn't in our memory map..
 181          * Fix it, but check if it's kernel or user first..
 182          */
 183
 184 bad_area:
 185         up_read(&mm->mmap_sem);
 186
 187 bad_area_nosemaphore:
 188
 189         /* User mode accesses just cause a SIGSEGV */
 190
 191         if (user_mode(regs)) {
 192                 info.si_signo = SIGSEGV;
 193                 info.si_errno = 0;
 194                 /* info.si_code has been set above */
 195                 info.si_addr = (void *)address;
 196                 force_sig_info(SIGSEGV, &info, tsk);
 197                 return;
 198         }
 199
 200 no_context:
 201
 202         /* Are we prepared to handle this kernel fault?
 203          *
 204          * (The kernel has valid exception-points in the source
 205          *  when it acesses user-memory. When it fails in one
 206          *  of those points, we find it in a table and do a jump
 207          *  to some fixup code that loads an appropriate error
 208          *  code)
 209          */
 210
 211         {
 212                 const struct exception_table_entry *entry;
 213
 214                 __asm__ __volatile__("l.nop 42");
 215
 216                 if ((entry = search_exception_tables(regs->pc)) != NULL) {
 217                         /* Adjust the instruction pointer in the stackframe */
 218                         regs->pc = entry->fixup;
 219                         return;
 220                 }
 221         }
 222
 223         /*
 224          * Oops. The kernel tried to access some bad page. We'll have to
 225          * terminate things with extreme prejudice.
 226          */
 227
 228         if ((unsigned long)(address) < PAGE_SIZE)
 229                 printk(KERN_ALERT
 230                        "Unable to handle kernel NULL pointer dereference");
 231         else
 232                 printk(KERN_ALERT "Unable to handle kernel access");
 233         printk(" at virtual address 0x%08lx\n", address);
 234
 235         die("Oops", regs, write_acc);
 236
 237         do_exit(SIGKILL);
 238
 239         /*
 240          * We ran out of memory, or some other thing happened to us that made
 241          * us unable to handle the page fault gracefully.
 242          */
 243
 244 out_of_memory:
 245         __asm__ __volatile__("l.nop 42");
 246         __asm__ __volatile__("l.nop 1");
 247
 248         up_read(&mm->mmap_sem);
 249         printk("VM: killing process %s\n", tsk->comm);
 250         if (user_mode(regs))
 251                 do_exit(SIGKILL);
 252         goto no_context;
 253
 254 do_sigbus:
 255         up_read(&mm->mmap_sem);
 256
 257         /*
 258          * Send a sigbus, regardless of whether we were in kernel
 259          * or user mode.
 260          */
 261         info.si_signo = SIGBUS;
 262         info.si_errno = 0;
 263         info.si_code = BUS_ADRERR;
 264         info.si_addr = (void *)address;
 265         force_sig_info(SIGBUS, &info, tsk);
 266
 267         /* Kernel mode? Handle exceptions or die */
 268         if (!user_mode(regs))
 269                 goto no_context;
 270         return;
 271
 272 vmalloc_fault:
 273         {
 274                 /*
 275                  * Synchronize this task's top level page-table
 276                  * with the 'reference' page table.
 277                  *
 278                  * Use current_pgd instead of tsk->active_mm->pgd
 279                  * since the latter might be unavailable if this
 280                  * code is executed in a misfortunately run irq
 281                  * (like inside schedule() between switch_mm and
 282                  *  switch_to...).
 283                  */
 284
 285                 int offset = pgd_index(address);
 286                 pgd_t *pgd, *pgd_k;
 287                 pud_t *pud, *pud_k;
 288                 pmd_t *pmd, *pmd_k;
 289                 pte_t *pte_k;
 290
 291 /*
 292                 phx_warn("do_page_fault(): vmalloc_fault will not work, "
 293                          "since current_pgd assign a proper value somewhere\n"
 294                          "anyhow we don't need this at the moment\n");
 295
 296                 phx_mmu("vmalloc_fault");
 297 */
 298                 pgd = (pgd_t *)current_pgd + offset;
 299                 pgd_k = init_mm.pgd + offset;
 300
 301                 /* Since we're two-level, we don't need to do both
 302                  * set_pgd and set_pmd (they do the same thing). If
 303                  * we go three-level at some point, do the right thing
 304                  * with pgd_present and set_pgd here.
 305                  *
 306                  * Also, since the vmalloc area is global, we don't
 307                  * need to copy individual PTE's, it is enough to
 308                  * copy the pgd pointer into the pte page of the
 309                  * root task. If that is there, we'll find our pte if
 310                  * it exists.
 311                  */
 312
 313                 pud = pud_offset(pgd, address);
 314                 pud_k = pud_offset(pgd_k, address);
 315                 if (!pud_present(*pud_k))
 316                         goto no_context;
 317
 318                 pmd = pmd_offset(pud, address);
 319                 pmd_k = pmd_offset(pud_k, address);
 320
 321                 if (!pmd_present(*pmd_k))
 322                         goto bad_area_nosemaphore;
 323
 324                 set_pmd(pmd, *pmd_k);
 325
 326                 /* Make sure the actual PTE exists as well to
 327                  * catch kernel vmalloc-area accesses to non-mapped
 328                  * addresses. If we don't do this, this will just
 329                  * silently loop forever.
 330                  */
 331
 332                 pte_k = pte_offset_kernel(pmd_k, address);
 333                 if (!pte_present(*pte_k))
 334                         goto no_context;
 335
 336                 return;
 337         }
 338 }