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/module.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/pgtable.h>
  34 #include <asm/mmu.h>
  35 #include <linux/mmu_context.h>
  36 #include <linux/uaccess.h>
  37 #include <asm/exceptions.h>
  38
  39 static unsigned long pte_misses;        /* updated by do_page_fault() */
  40 static unsigned long pte_errors;        /* updated by do_page_fault() */
  41
  42 /*
  43  * Check whether the instruction at regs->pc is a store using
  44  * an update addressing form which will update r1.
  45  */
  46 static int store_updates_sp(struct pt_regs *regs)
  47 {
  48         unsigned int inst;
  49
  50         if (get_user(inst, (unsigned int __user *)regs->pc))
  51                 return 0;
  52         /* check for 1 in the rD field */
  53         if (((inst >> 21) & 0x1f) != 1)
  54                 return 0;
  55         /* check for store opcodes */
  56         if ((inst & 0xd0000000) == 0xd0000000)
  57                 return 1;
  58         return 0;
  59 }
  60
  61
  62 /*
  63  * bad_page_fault is called when we have a bad access from the kernel.
  64  * It is called from do_page_fault above and from some of the procedures
  65  * in traps.c.
  66  */
  67 void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
  68 {
  69         const struct exception_table_entry *fixup;
  70 /* MS: no context */
  71         /* Are we prepared to handle this fault?  */
  72         fixup = search_exception_tables(regs->pc);
  73         if (fixup) {
  74                 regs->pc = fixup->fixup;
  75                 return;
  76         }
  77
  78         /* kernel has accessed a bad area */
  79         die("kernel access of bad area", regs, sig);
  80 }
  81
  82 /*
  83  * The error_code parameter is ESR for a data fault,
  84  * 0 for an instruction fault.
  85  */
  86 void do_page_fault(struct pt_regs *regs, unsigned long address,
  87                    unsigned long error_code)
  88 {
  89         struct vm_area_struct *vma;
  90         struct mm_struct *mm = current->mm;
  91         siginfo_t info;
  92         int code = SEGV_MAPERR;
  93         int is_write = error_code & ESR_S;
  94         int fault;
  95         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
  96                                          (is_write ? FAULT_FLAG_WRITE : 0);
  97
  98         regs->ear = address;
  99         regs->esr = error_code;
 100
 101         /* On a kernel SLB miss we can only check for a valid exception entry */
 102         if (unlikely(kernel_mode(regs) && (address >= TASK_SIZE))) {
 103                 pr_warn("kernel task_size exceed");
 104                 _exception(SIGSEGV, regs, code, address);
 105         }
 106
 107         /* for instr TLB miss and instr storage exception ESR_S is undefined */
 108         if ((error_code & 0x13) == 0x13 || (error_code & 0x11) == 0x11)
 109                 is_write = 0;
 110
 111         if (unlikely(in_atomic() || !mm)) {
 112                 if (kernel_mode(regs))
 113                         goto bad_area_nosemaphore;
 114
 115                 /* in_atomic() in user mode is really bad,
 116                    as is current->mm == NULL. */
 117                 pr_emerg("Page fault in user mode with in_atomic(), mm = %p\n",
 118                                                                         mm);
 119                 pr_emerg("r15 = %lx  MSR = %lx\n",
 120                        regs->r15, regs->msr);
 121                 die("Weird page fault", regs, SIGSEGV);
 122         }
 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(!down_read_trylock(&mm->mmap_sem))) {
 140                 if (kernel_mode(regs) && !search_exception_tables(regs->pc))
 141                         goto bad_area_nosemaphore;
 142
 143 retry:
 144                 down_read(&mm->mmap_sem);
 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         /* a read */
 203         } else {
 204                 /* protection fault */
 205                 if (unlikely(error_code & 0x08000000))
 206                         goto bad_area;
 207                 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC))))
 208                         goto bad_area;
 209         }
 210
 211         /*
 212          * If for any reason at all we couldn't handle the fault,
 213          * make sure we exit gracefully rather than endlessly redo
 214          * the fault.
 215          */
 216         fault = handle_mm_fault(mm, vma, address, flags);
 217
 218         if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
 219                 return;
 220
 221         if (unlikely(fault & VM_FAULT_ERROR)) {
 222                 if (fault & VM_FAULT_OOM)
 223                         goto out_of_memory;
 224                 else if (fault & VM_FAULT_SIGBUS)
 225                         goto do_sigbus;
 226                 BUG();
 227         }
 228
 229         if (flags & FAULT_FLAG_ALLOW_RETRY) {
 230                 if (unlikely(fault & VM_FAULT_MAJOR))
 231                         current->maj_flt++;
 232                 else
 233                         current->min_flt++;
 234                 if (fault & VM_FAULT_RETRY) {
 235                         flags &= ~FAULT_FLAG_ALLOW_RETRY;
 236                         flags |= FAULT_FLAG_TRIED;
 237
 238                         /*
 239                          * No need to up_read(&mm->mmap_sem) as we would
 240                          * have already released it in __lock_page_or_retry
 241                          * in mm/filemap.c.
 242                          */
 243
 244                         goto retry;
 245                 }
 246         }
 247
 248         up_read(&mm->mmap_sem);
 249
 250         /*
 251          * keep track of tlb+htab misses that are good addrs but
 252          * just need pte's created via handle_mm_fault()
 253          * -- Cort
 254          */
 255         pte_misses++;
 256         return;
 257
 258 bad_area:
 259         up_read(&mm->mmap_sem);
 260
 261 bad_area_nosemaphore:
 262         pte_errors++;
 263
 264         /* User mode accesses cause a SIGSEGV */
 265         if (user_mode(regs)) {
 266                 _exception(SIGSEGV, regs, code, address);
 267 /*              info.si_signo = SIGSEGV;
 268                 info.si_errno = 0;
 269                 info.si_code = code;
 270                 info.si_addr = (void *) address;
 271                 force_sig_info(SIGSEGV, &info, current);*/
 272                 return;
 273         }
 274
 275         bad_page_fault(regs, address, SIGSEGV);
 276         return;
 277
 278 /*
 279  * We ran out of memory, or some other thing happened to us that made
 280  * us unable to handle the page fault gracefully.
 281  */
 282 out_of_memory:
 283         up_read(&mm->mmap_sem);
 284         if (!user_mode(regs))
 285                 bad_page_fault(regs, address, SIGKILL);
 286         else
 287                 pagefault_out_of_memory();
 288         return;
 289
 290 do_sigbus:
 291         up_read(&mm->mmap_sem);
 292         if (user_mode(regs)) {
 293                 info.si_signo = SIGBUS;
 294                 info.si_errno = 0;
 295                 info.si_code = BUS_ADRERR;
 296                 info.si_addr = (void __user *)address;
 297                 force_sig_info(SIGBUS, &info, current);
 298                 return;
 299         }
 300         bad_page_fault(regs, address, SIGBUS);
 301 }