arch/microblaze/mm/fault.c

   1 /*
   2  *  arch/microblaze/mm/fault.c
   3  *
   4  *    Copyright (C) 2007 Xilinx, Inc.  All rights reserved.
   5  *
   6  *  Derived from "arch/ppc/mm/fault.c"
   7  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
   8  *
   9  *  Derived from "arch/i386/mm/fault.c"
  10  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
  11  *
  12  *  Modified by Cort Dougan and Paul Mackerras.
  13  *
  14  * This file is subject to the terms and conditions of the GNU General
  15  * Public License.  See the file COPYING in the main directory of this
  16  * archive for more details.
  17  *
  18  */
  19
  20 #include <linux/extable.h>
  21 #include <linux/signal.h>
  22 #include <linux/sched.h>
  23 #include <linux/kernel.h>
  24 #include <linux/errno.h>
  25 #include <linux/string.h>
  26 #include <linux/types.h>
  27 #include <linux/ptrace.h>
  28 #include <linux/mman.h>
  29 #include <linux/mm.h>
  30 #include <linux/interrupt.h>
  31
  32 #include <asm/page.h>
  33 #include <asm/mmu.h>
  34 #include <linux/mmu_context.h>
  35 #include <linux/uaccess.h>
  36 #include <asm/exceptions.h>
  37
  38 static unsigned long pte_misses;        /* updated by do_page_fault() */
  39 static unsigned long pte_errors;        /* updated by do_page_fault() */
  40
  41 /*
  42  * Check whether the instruction at regs->pc is a store using
  43  * an update addressing form which will update r1.
  44  */
  45 static int store_updates_sp(struct pt_regs *regs)
  46 {
  47         unsigned int inst;
  48
  49         if (get_user(inst, (unsigned int __user *)regs->pc))
  50                 return 0;
  51         /* check for 1 in the rD field */
  52         if (((inst >> 21) & 0x1f) != 1)
  53                 return 0;
  54         /* check for store opcodes */
  55         if ((inst & 0xd0000000) == 0xd0000000)
  56                 return 1;
  57         return 0;
  58 }
  59
  60
  61 /*
  62  * bad_page_fault is called when we have a bad access from the kernel.
  63  * It is called from do_page_fault above and from some of the procedures
  64  * in traps.c.
  65  */
  66 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
  67 {
  68         const struct exception_table_entry *fixup;
  69 /* MS: no context */
  70         /* Are we prepared to handle this fault?  */
  71         fixup = search_exception_tables(regs->pc);
  72         if (fixup) {
  73                 regs->pc = fixup->fixup;
  74                 return;
  75         }
  76
  77         /* kernel has accessed a bad area */
  78         die("kernel access of bad area", regs, sig);
  79 }
  80
  81 /*
  82  * The error_code parameter is ESR for a data fault,
  83  * 0 for an instruction fault.
  84  */
  85 void do_page_fault(struct pt_regs *regs, unsigned long address,
  86                    unsigned long error_code)
  87 {
  88         struct vm_area_struct *vma;
  89         struct mm_struct *mm = current->mm;
  90         int code = SEGV_MAPERR;
  91         int is_write = error_code & ESR_S;
  92         vm_fault_t fault;
  93         unsigned int flags = FAULT_FLAG_DEFAULT;
  94
  95         regs->ear = address;
  96         regs->esr = error_code;
  97
  98         /* On a kernel SLB miss we can only check for a valid exception entry */
  99         if (unlikely(kernel_mode(regs) && (address >= TASK_SIZE))) {
 100                 pr_warn("kernel task_size exceed");
 101                 _exception(SIGSEGV, regs, code, address);
 102         }
 103
 104         /* for instr TLB miss and instr storage exception ESR_S is undefined */
 105         if ((error_code & 0x13) == 0x13 || (error_code & 0x11) == 0x11)
 106                 is_write = 0;
 107
 108         if (unlikely(faulthandler_disabled() || !mm)) {
 109                 if (kernel_mode(regs))
 110                         goto bad_area_nosemaphore;
 111
 112                 /* faulthandler_disabled() in user mode is really bad,
 113                    as is current->mm == NULL. */
 114                 pr_emerg("Page fault in user mode with faulthandler_disabled(), mm = %p\n",
 115                          mm);
 116                 pr_emerg("r15 = %lx  MSR = %lx\n",
 117                        regs->r15, regs->msr);
 118                 die("Weird page fault", regs, SIGSEGV);
 119         }
 120
 121         if (user_mode(regs))
 122                 flags |= FAULT_FLAG_USER;
 123
 124         /* When running in the kernel we expect faults to occur only to
 125          * addresses in user space.  All other faults represent errors in the
 126          * kernel and should generate an OOPS.  Unfortunately, in the case of an
 127          * erroneous fault occurring in a code path which already holds mmap_sem
 128          * we will deadlock attempting to validate the fault against the
 129          * address space.  Luckily the kernel only validly references user
 130          * space from well defined areas of code, which are listed in the
 131          * exceptions table.
 132          *
 133          * As the vast majority of faults will be valid we will only perform
 134          * the source reference check when there is a possibility of a deadlock.
 135          * Attempt to lock the address space, if we cannot we then validate the
 136          * source.  If this is invalid we can skip the address space check,
 137          * thus avoiding the deadlock.
 138          */
 139         if (unlikely(!mmap_read_trylock(mm))) {
 140                 if (kernel_mode(regs) && !search_exception_tables(regs->pc))
 141                         goto bad_area_nosemaphore;
 142
 143 retry:
 144                 mmap_read_lock(mm);
 145         }
 146
 147         vma = find_vma(mm, address);
 148         if (unlikely(!vma))
 149                 goto bad_area;
 150
 151         if (vma->vm_start <= address)
 152                 goto good_area;
 153
 154         if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
 155                 goto bad_area;
 156
 157         if (unlikely(!is_write))
 158                 goto bad_area;
 159
 160         /*
 161          * N.B. The ABI allows programs to access up to
 162          * a few hundred bytes below the stack pointer (TBD).
 163          * The kernel signal delivery code writes up to about 1.5kB
 164          * below the stack pointer (r1) before decrementing it.
 165          * The exec code can write slightly over 640kB to the stack
 166          * before setting the user r1.  Thus we allow the stack to
 167          * expand to 1MB without further checks.
 168          */
 169         if (unlikely(address + 0x100000 < vma->vm_end)) {
 170
 171                 /* get user regs even if this fault is in kernel mode */
 172                 struct pt_regs *uregs = current->thread.regs;
 173                 if (uregs == NULL)
 174                         goto bad_area;
 175
 176                 /*
 177                  * A user-mode access to an address a long way below
 178                  * the stack pointer is only valid if the instruction
 179                  * is one which would update the stack pointer to the
 180                  * address accessed if the instruction completed,
 181                  * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
 182                  * (or the byte, halfword, float or double forms).
 183                  *
 184                  * If we don't check this then any write to the area
 185                  * between the last mapped region and the stack will
 186                  * expand the stack rather than segfaulting.
 187                  */
 188                 if (address + 2048 < uregs->r1
 189                         && (kernel_mode(regs) || !store_updates_sp(regs)))
 190                                 goto bad_area;
 191         }
 192         if (expand_stack(vma, address))
 193                 goto bad_area;
 194
 195 good_area:
 196         code = SEGV_ACCERR;
 197
 198         /* a write */
 199         if (unlikely(is_write)) {
 200                 if (unlikely(!(vma->vm_flags & VM_WRITE)))
 201                         goto bad_area;
 202                 flags |= FAULT_FLAG_WRITE;
 203         /* a read */
 204         } else {
 205                 /* protection fault */
 206                 if (unlikely(error_code & 0x08000000))
 207                         goto bad_area;
 208                 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC))))
 209                         goto bad_area;
 210         }
 211
 212         /*
 213          * If for any reason at all we couldn't handle the fault,
 214          * make sure we exit gracefully rather than endlessly redo
 215          * the fault.
 216          */
 217         fault = handle_mm_fault(vma, address, flags);
 218
 219         if (fault_signal_pending(fault, regs))
 220                 return;
 221
 222         if (unlikely(fault & VM_FAULT_ERROR)) {
 223                 if (fault & VM_FAULT_OOM)
 224                         goto out_of_memory;
 225                 else if (fault & VM_FAULT_SIGSEGV)
 226                         goto bad_area;
 227                 else if (fault & VM_FAULT_SIGBUS)
 228                         goto do_sigbus;
 229                 BUG();
 230         }
 231
 232         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 233                 if (unlikely(fault & VM_FAULT_MAJOR))
 234                         current->maj_flt++;
 235                 else
 236                         current->min_flt++;
 237                 if (fault & VM_FAULT_RETRY) {
 238                         flags |= FAULT_FLAG_TRIED;
 239
 240                         /*
 241                          * No need to up_read(&mm->mmap_sem) as we would
 242                          * have already released it in __lock_page_or_retry
 243                          * in mm/filemap.c.
 244                          */
 245
 246                         goto retry;
 247                 }
 248         }
 249
 250         mmap_read_unlock(mm);
 251
 252         /*
 253          * keep track of tlb+htab misses that are good addrs but
 254          * just need pte's created via handle_mm_fault()
 255          * -- Cort
 256          */
 257         pte_misses++;
 258         return;
 259
 260 bad_area:
 261         mmap_read_unlock(mm);
 262
 263 bad_area_nosemaphore:
 264         pte_errors++;
 265
 266         /* User mode accesses cause a SIGSEGV */
 267         if (user_mode(regs)) {
 268                 _exception(SIGSEGV, regs, code, address);
 269                 return;
 270         }
 271
 272         bad_page_fault(regs, address, SIGSEGV);
 273         return;
 274
 275 /*
 276  * We ran out of memory, or some other thing happened to us that made
 277  * us unable to handle the page fault gracefully.
 278  */
 279 out_of_memory:
 280         mmap_read_unlock(mm);
 281         if (!user_mode(regs))
 282                 bad_page_fault(regs, address, SIGKILL);
 283         else
 284                 pagefault_out_of_memory();
 285         return;
 286
 287 do_sigbus:
 288         mmap_read_unlock(mm);
 289         if (user_mode(regs)) {
 290                 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
 291                 return;
 292         }
 293         bad_page_fault(regs, address, SIGBUS);
 294 }