arch/sh/mm/fault.c

   1 /*
   2  * Page fault handler for SH with an MMU.
   3  *
   4  *  Copyright (C) 1999  Niibe Yutaka
   5  *  Copyright (C) 2003 - 2007  Paul Mundt
   6  *
   7  *  Based on linux/arch/i386/mm/fault.c:
   8  *   Copyright (C) 1995  Linus Torvalds
   9  *
  10  * This file is subject to the terms and conditions of the GNU General Public
  11  * License.  See the file "COPYING" in the main directory of this archive
  12  * for more details.
  13  */
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/hardirq.h>
  17 #include <linux/kprobes.h>
  18 #include <asm/system.h>
  19 #include <asm/mmu_context.h>
  20 #include <asm/tlbflush.h>
  21 #include <asm/kgdb.h>
  22
  23 /*
  24  * This routine handles page faults.  It determines the address,
  25  * and the problem, and then passes it off to one of the appropriate
  26  * routines.
  27  */
  28 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
  29                                         unsigned long writeaccess,
  30                                         unsigned long address)
  31 {
  32         struct task_struct *tsk;
  33         struct mm_struct *mm;
  34         struct vm_area_struct * vma;
  35         int si_code;
  36         siginfo_t info;
  37
  38         trace_hardirqs_on();
  39         local_irq_enable();
  40
  41 #ifdef CONFIG_SH_KGDB
  42         if (kgdb_nofault && kgdb_bus_err_hook)
  43                 kgdb_bus_err_hook();
  44 #endif
  45
  46         tsk = current;
  47         mm = tsk->mm;
  48         si_code = SEGV_MAPERR;
  49
  50         if (unlikely(address >= TASK_SIZE)) {
  51                 /*
  52                  * Synchronize this task's top level page-table
  53                  * with the 'reference' page table.
  54                  *
  55                  * Do _not_ use "tsk" here. We might be inside
  56                  * an interrupt in the middle of a task switch..
  57                  */
  58                 int offset = pgd_index(address);
  59                 pgd_t *pgd, *pgd_k;
  60                 pud_t *pud, *pud_k;
  61                 pmd_t *pmd, *pmd_k;
  62
  63                 pgd = get_TTB() + offset;
  64                 pgd_k = swapper_pg_dir + offset;
  65
  66                 /* This will never happen with the folded page table. */
  67                 if (!pgd_present(*pgd)) {
  68                         if (!pgd_present(*pgd_k))
  69                                 goto bad_area_nosemaphore;
  70                         set_pgd(pgd, *pgd_k);
  71                         return;
  72                 }
  73
  74                 pud = pud_offset(pgd, address);
  75                 pud_k = pud_offset(pgd_k, address);
  76                 if (pud_present(*pud) || !pud_present(*pud_k))
  77                         goto bad_area_nosemaphore;
  78                 set_pud(pud, *pud_k);
  79
  80                 pmd = pmd_offset(pud, address);
  81                 pmd_k = pmd_offset(pud_k, address);
  82                 if (pmd_present(*pmd) || !pmd_present(*pmd_k))
  83                         goto bad_area_nosemaphore;
  84                 set_pmd(pmd, *pmd_k);
  85
  86                 return;
  87         }
  88
  89         /*
  90          * If we're in an interrupt or have no user
  91          * context, we must not take the fault..
  92          */
  93         if (in_atomic() || !mm)
  94                 goto no_context;
  95
  96         down_read(&mm->mmap_sem);
  97
  98         vma = find_vma(mm, address);
  99         if (!vma)
 100                 goto bad_area;
 101         if (vma->vm_start <= address)
 102                 goto good_area;
 103         if (!(vma->vm_flags & VM_GROWSDOWN))
 104                 goto bad_area;
 105         if (expand_stack(vma, address))
 106                 goto bad_area;
 107 /*
 108  * Ok, we have a good vm_area for this memory access, so
 109  * we can handle it..
 110  */
 111 good_area:
 112         si_code = SEGV_ACCERR;
 113         if (writeaccess) {
 114                 if (!(vma->vm_flags & VM_WRITE))
 115                         goto bad_area;
 116         } else {
 117                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 118                         goto bad_area;
 119         }
 120
 121         /*
 122          * If for any reason at all we couldn't handle the fault,
 123          * make sure we exit gracefully rather than endlessly redo
 124          * the fault.
 125          */
 126 survive:
 127         switch (handle_mm_fault(mm, vma, address, writeaccess)) {
 128                 case VM_FAULT_MINOR:
 129                         tsk->min_flt++;
 130                         break;
 131                 case VM_FAULT_MAJOR:
 132                         tsk->maj_flt++;
 133                         break;
 134                 case VM_FAULT_SIGBUS:
 135                         goto do_sigbus;
 136                 case VM_FAULT_OOM:
 137                         goto out_of_memory;
 138                 default:
 139                         BUG();
 140         }
 141
 142         up_read(&mm->mmap_sem);
 143         return;
 144
 145 /*
 146  * Something tried to access memory that isn't in our memory map..
 147  * Fix it, but check if it's kernel or user first..
 148  */
 149 bad_area:
 150         up_read(&mm->mmap_sem);
 151
 152 bad_area_nosemaphore:
 153         if (user_mode(regs)) {
 154                 info.si_signo = SIGSEGV;
 155                 info.si_errno = 0;
 156                 info.si_code = si_code;
 157                 info.si_addr = (void *) address;
 158                 force_sig_info(SIGSEGV, &info, tsk);
 159                 return;
 160         }
 161
 162 no_context:
 163         /* Are we prepared to handle this kernel fault?  */
 164         if (fixup_exception(regs))
 165                 return;
 166
 167 /*
 168  * Oops. The kernel tried to access some bad page. We'll have to
 169  * terminate things with extreme prejudice.
 170  *
 171  */
 172
 173         bust_spinlocks(1);
 174
 175         if (oops_may_print()) {
 176                 __typeof__(pte_val(__pte(0))) page;
 177
 178                 if (address < PAGE_SIZE)
 179                         printk(KERN_ALERT "Unable to handle kernel NULL "
 180                                           "pointer dereference");
 181                 else
 182                         printk(KERN_ALERT "Unable to handle kernel paging "
 183                                           "request");
 184                 printk(" at virtual address %08lx\n", address);
 185                 printk(KERN_ALERT "pc = %08lx\n", regs->pc);
 186                 page = (unsigned long)get_TTB();
 187                 if (page) {
 188                         page = ((__typeof__(page) *) __va(page))[address >>
 189                                                                  PGDIR_SHIFT];
 190                         printk(KERN_ALERT "*pde = %08lx\n", page);
 191                         if (page & _PAGE_PRESENT) {
 192                                 page &= PAGE_MASK;
 193                                 address &= 0x003ff000;
 194                                 page = ((__typeof__(page) *)
 195                                                 __va(page))[address >>
 196                                                             PAGE_SHIFT];
 197                                 printk(KERN_ALERT "*pte = %08lx\n", page);
 198                         }
 199                 }
 200         }
 201
 202         die("Oops", regs, writeaccess);
 203         bust_spinlocks(0);
 204         do_exit(SIGKILL);
 205
 206 /*
 207  * We ran out of memory, or some other thing happened to us that made
 208  * us unable to handle the page fault gracefully.
 209  */
 210 out_of_memory:
 211         up_read(&mm->mmap_sem);
 212         if (is_init(current)) {
 213                 yield();
 214                 down_read(&mm->mmap_sem);
 215                 goto survive;
 216         }
 217         printk("VM: killing process %s\n", tsk->comm);
 218         if (user_mode(regs))
 219                 do_exit(SIGKILL);
 220         goto no_context;
 221
 222 do_sigbus:
 223         up_read(&mm->mmap_sem);
 224
 225         /*
 226          * Send a sigbus, regardless of whether we were in kernel
 227          * or user mode.
 228          */
 229         info.si_signo = SIGBUS;
 230         info.si_errno = 0;
 231         info.si_code = BUS_ADRERR;
 232         info.si_addr = (void *)address;
 233         force_sig_info(SIGBUS, &info, tsk);
 234
 235         /* Kernel mode? Handle exceptions or die */
 236         if (!user_mode(regs))
 237                 goto no_context;
 238 }
 239
 240 #ifdef CONFIG_SH_STORE_QUEUES
 241 /*
 242  * This is a special case for the SH-4 store queues, as pages for this
 243  * space still need to be faulted in before it's possible to flush the
 244  * store queue cache for writeout to the remapped region.
 245  */
 246 #define P3_ADDR_MAX             (P4SEG_STORE_QUE + 0x04000000)
 247 #else
 248 #define P3_ADDR_MAX             P4SEG
 249 #endif
 250
 251 /*
 252  * Called with interrupts disabled.
 253  */
 254 asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
 255                                          unsigned long writeaccess,
 256                                          unsigned long address)
 257 {
 258         pgd_t *pgd;
 259         pud_t *pud;
 260         pmd_t *pmd;
 261         pte_t *pte;
 262         pte_t entry;
 263         struct mm_struct *mm = current->mm;
 264         spinlock_t *ptl = NULL;
 265         int ret = 1;
 266
 267 #ifdef CONFIG_SH_KGDB
 268         if (kgdb_nofault && kgdb_bus_err_hook)
 269                 kgdb_bus_err_hook();
 270 #endif
 271
 272         /*
 273          * We don't take page faults for P1, P2, and parts of P4, these
 274          * are always mapped, whether it be due to legacy behaviour in
 275          * 29-bit mode, or due to PMB configuration in 32-bit mode.
 276          */
 277         if (address >= P3SEG && address < P3_ADDR_MAX) {
 278                 pgd = pgd_offset_k(address);
 279                 mm = NULL;
 280         } else {
 281                 if (unlikely(address >= TASK_SIZE || !mm))
 282                         return 1;
 283
 284                 pgd = pgd_offset(mm, address);
 285         }
 286
 287         pud = pud_offset(pgd, address);
 288         if (pud_none_or_clear_bad(pud))
 289                 return 1;
 290         pmd = pmd_offset(pud, address);
 291         if (pmd_none_or_clear_bad(pmd))
 292                 return 1;
 293
 294         if (mm)
 295                 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
 296         else
 297                 pte = pte_offset_kernel(pmd, address);
 298
 299         entry = *pte;
 300         if (unlikely(pte_none(entry) || pte_not_present(entry)))
 301                 goto unlock;
 302         if (unlikely(writeaccess && !pte_write(entry)))
 303                 goto unlock;
 304
 305         if (writeaccess)
 306                 entry = pte_mkdirty(entry);
 307         entry = pte_mkyoung(entry);
 308
 309 #ifdef CONFIG_CPU_SH4
 310         /*
 311          * ITLB is not affected by "ldtlb" instruction.
 312          * So, we need to flush the entry by ourselves.
 313          */
 314         local_flush_tlb_one(get_asid(), address & PAGE_MASK);
 315 #endif
 316
 317         set_pte(pte, entry);
 318         update_mmu_cache(NULL, address, entry);
 319         ret = 0;
 320 unlock:
 321         if (mm)
 322                 pte_unmap_unlock(pte, ptl);
 323         return ret;
 324 }