arch/sh/mm/fault_32.c

   1 /*
   2  * Page fault handler for SH with an MMU.
   3  *
   4  *  Copyright (C) 1999  Niibe Yutaka
   5  *  Copyright (C) 2003 - 2009  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 <linux/perf_event.h>
  19 #include <asm/io_trapped.h>
  20 #include <asm/system.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/tlbflush.h>
  23
  24 static inline int notify_page_fault(struct pt_regs *regs, int trap)
  25 {
  26         int ret = 0;
  27
  28         if (kprobes_built_in() && !user_mode(regs)) {
  29                 preempt_disable();
  30                 if (kprobe_running() && kprobe_fault_handler(regs, trap))
  31                         ret = 1;
  32                 preempt_enable();
  33         }
  34
  35         return ret;
  36 }
  37
  38 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
  39 {
  40         unsigned index = pgd_index(address);
  41         pgd_t *pgd_k;
  42         pud_t *pud, *pud_k;
  43         pmd_t *pmd, *pmd_k;
  44
  45         pgd += index;
  46         pgd_k = init_mm.pgd + index;
  47
  48         if (!pgd_present(*pgd_k))
  49                 return NULL;
  50
  51         pud = pud_offset(pgd, address);
  52         pud_k = pud_offset(pgd_k, address);
  53         if (!pud_present(*pud_k))
  54                 return NULL;
  55
  56         pmd = pmd_offset(pud, address);
  57         pmd_k = pmd_offset(pud_k, address);
  58         if (!pmd_present(*pmd_k))
  59                 return NULL;
  60
  61         if (!pmd_present(*pmd))
  62                 set_pmd(pmd, *pmd_k);
  63         else {
  64                 /*
  65                  * The page tables are fully synchronised so there must
  66                  * be another reason for the fault. Return NULL here to
  67                  * signal that we have not taken care of the fault.
  68                  */
  69                 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
  70                 return NULL;
  71         }
  72
  73         return pmd_k;
  74 }
  75
  76 /*
  77  * Handle a fault on the vmalloc or module mapping area
  78  */
  79 static noinline int vmalloc_fault(unsigned long address)
  80 {
  81         pgd_t *pgd_k;
  82         pmd_t *pmd_k;
  83         pte_t *pte_k;
  84
  85         /* Make sure we are in vmalloc/module/P3 area: */
  86         if (!(address >= VMALLOC_START && address < P3_ADDR_MAX))
  87                 return -1;
  88
  89         /*
  90          * Synchronize this task's top level page-table
  91          * with the 'reference' page table.
  92          *
  93          * Do _not_ use "current" here. We might be inside
  94          * an interrupt in the middle of a task switch..
  95          */
  96         pgd_k = get_TTB();
  97         pmd_k = vmalloc_sync_one(pgd_k, address);
  98         if (!pmd_k)
  99                 return -1;
 100
 101         pte_k = pte_offset_kernel(pmd_k, address);
 102         if (!pte_present(*pte_k))
 103                 return -1;
 104
 105         return 0;
 106 }
 107
 108 static int fault_in_kernel_space(unsigned long address)
 109 {
 110         return address >= TASK_SIZE;
 111 }
 112
 113 /*
 114  * This routine handles page faults.  It determines the address,
 115  * and the problem, and then passes it off to one of the appropriate
 116  * routines.
 117  */
 118 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 119                                         unsigned long writeaccess,
 120                                         unsigned long address)
 121 {
 122         unsigned long vec;
 123         struct task_struct *tsk;
 124         struct mm_struct *mm;
 125         struct vm_area_struct * vma;
 126         int si_code;
 127         int fault;
 128         siginfo_t info;
 129
 130         tsk = current;
 131         mm = tsk->mm;
 132         si_code = SEGV_MAPERR;
 133         vec = lookup_exception_vector();
 134
 135         /*
 136          * We fault-in kernel-space virtual memory on-demand. The
 137          * 'reference' page table is init_mm.pgd.
 138          *
 139          * NOTE! We MUST NOT take any locks for this case. We may
 140          * be in an interrupt or a critical region, and should
 141          * only copy the information from the master page table,
 142          * nothing more.
 143          */
 144         if (unlikely(fault_in_kernel_space(address))) {
 145                 if (vmalloc_fault(address) >= 0)
 146                         return;
 147                 if (notify_page_fault(regs, vec))
 148                         return;
 149
 150                 goto bad_area_nosemaphore;
 151         }
 152
 153         if (unlikely(notify_page_fault(regs, vec)))
 154                 return;
 155
 156         /* Only enable interrupts if they were on before the fault */
 157         if ((regs->sr & SR_IMASK) != SR_IMASK)
 158                 local_irq_enable();
 159
 160         perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
 161
 162         /*
 163          * If we're in an interrupt, have no user context or are running
 164          * in an atomic region then we must not take the fault:
 165          */
 166         if (in_atomic() || !mm)
 167                 goto no_context;
 168
 169         down_read(&mm->mmap_sem);
 170
 171         vma = find_vma(mm, address);
 172         if (!vma)
 173                 goto bad_area;
 174         if (vma->vm_start <= address)
 175                 goto good_area;
 176         if (!(vma->vm_flags & VM_GROWSDOWN))
 177                 goto bad_area;
 178         if (expand_stack(vma, address))
 179                 goto bad_area;
 180
 181         /*
 182          * Ok, we have a good vm_area for this memory access, so
 183          * we can handle it..
 184          */
 185 good_area:
 186         si_code = SEGV_ACCERR;
 187         if (writeaccess) {
 188                 if (!(vma->vm_flags & VM_WRITE))
 189                         goto bad_area;
 190         } else {
 191                 if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
 192                         goto bad_area;
 193         }
 194
 195         /*
 196          * If for any reason at all we couldn't handle the fault,
 197          * make sure we exit gracefully rather than endlessly redo
 198          * the fault.
 199          */
 200 survive:
 201         fault = handle_mm_fault(mm, vma, address, writeaccess ? FAULT_FLAG_WRITE : 0);
 202         if (unlikely(fault & VM_FAULT_ERROR)) {
 203                 if (fault & VM_FAULT_OOM)
 204                         goto out_of_memory;
 205                 else if (fault & VM_FAULT_SIGBUS)
 206                         goto do_sigbus;
 207                 BUG();
 208         }
 209         if (fault & VM_FAULT_MAJOR) {
 210                 tsk->maj_flt++;
 211                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
 212                                      regs, address);
 213         } else {
 214                 tsk->min_flt++;
 215                 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
 216                                      regs, address);
 217         }
 218
 219         up_read(&mm->mmap_sem);
 220         return;
 221
 222         /*
 223          * Something tried to access memory that isn't in our memory map..
 224          * Fix it, but check if it's kernel or user first..
 225          */
 226 bad_area:
 227         up_read(&mm->mmap_sem);
 228
 229 bad_area_nosemaphore:
 230         if (user_mode(regs)) {
 231                 info.si_signo = SIGSEGV;
 232                 info.si_errno = 0;
 233                 info.si_code = si_code;
 234                 info.si_addr = (void *) address;
 235                 force_sig_info(SIGSEGV, &info, tsk);
 236                 return;
 237         }
 238
 239 no_context:
 240         /* Are we prepared to handle this kernel fault?  */
 241         if (fixup_exception(regs))
 242                 return;
 243
 244         if (handle_trapped_io(regs, address))
 245                 return;
 246 /*
 247  * Oops. The kernel tried to access some bad page. We'll have to
 248  * terminate things with extreme prejudice.
 249  *
 250  */
 251
 252         bust_spinlocks(1);
 253
 254         if (oops_may_print()) {
 255                 unsigned long page;
 256
 257                 if (address < PAGE_SIZE)
 258                         printk(KERN_ALERT "Unable to handle kernel NULL "
 259                                           "pointer dereference");
 260                 else
 261                         printk(KERN_ALERT "Unable to handle kernel paging "
 262                                           "request");
 263                 printk(" at virtual address %08lx\n", address);
 264                 printk(KERN_ALERT "pc = %08lx\n", regs->pc);
 265                 page = (unsigned long)get_TTB();
 266                 if (page) {
 267                         page = ((__typeof__(page) *)page)[address >> PGDIR_SHIFT];
 268                         printk(KERN_ALERT "*pde = %08lx\n", page);
 269                         if (page & _PAGE_PRESENT) {
 270                                 page &= PAGE_MASK;
 271                                 address &= 0x003ff000;
 272                                 page = ((__typeof__(page) *)
 273                                                 __va(page))[address >>
 274                                                             PAGE_SHIFT];
 275                                 printk(KERN_ALERT "*pte = %08lx\n", page);
 276                         }
 277                 }
 278         }
 279
 280         die("Oops", regs, writeaccess);
 281         bust_spinlocks(0);
 282         do_exit(SIGKILL);
 283
 284 /*
 285  * We ran out of memory, or some other thing happened to us that made
 286  * us unable to handle the page fault gracefully.
 287  */
 288 out_of_memory:
 289         up_read(&mm->mmap_sem);
 290         if (is_global_init(current)) {
 291                 yield();
 292                 down_read(&mm->mmap_sem);
 293                 goto survive;
 294         }
 295         printk("VM: killing process %s\n", tsk->comm);
 296         if (user_mode(regs))
 297                 do_group_exit(SIGKILL);
 298         goto no_context;
 299
 300 do_sigbus:
 301         up_read(&mm->mmap_sem);
 302
 303         /*
 304          * Send a sigbus, regardless of whether we were in kernel
 305          * or user mode.
 306          */
 307         info.si_signo = SIGBUS;
 308         info.si_errno = 0;
 309         info.si_code = BUS_ADRERR;
 310         info.si_addr = (void *)address;
 311         force_sig_info(SIGBUS, &info, tsk);
 312
 313         /* Kernel mode? Handle exceptions or die */
 314         if (!user_mode(regs))
 315                 goto no_context;
 316 }
 317
 318 /*
 319  * Called with interrupts disabled.
 320  */
 321 asmlinkage int __kprobes
 322 handle_tlbmiss(struct pt_regs *regs, unsigned long writeaccess,
 323                unsigned long address)
 324 {
 325         pgd_t *pgd;
 326         pud_t *pud;
 327         pmd_t *pmd;
 328         pte_t *pte;
 329         pte_t entry;
 330
 331         /*
 332          * We don't take page faults for P1, P2, and parts of P4, these
 333          * are always mapped, whether it be due to legacy behaviour in
 334          * 29-bit mode, or due to PMB configuration in 32-bit mode.
 335          */
 336         if (address >= P3SEG && address < P3_ADDR_MAX) {
 337                 pgd = pgd_offset_k(address);
 338         } else {
 339                 if (unlikely(address >= TASK_SIZE || !current->mm))
 340                         return 1;
 341
 342                 pgd = pgd_offset(current->mm, address);
 343         }
 344
 345         pud = pud_offset(pgd, address);
 346         if (pud_none_or_clear_bad(pud))
 347                 return 1;
 348         pmd = pmd_offset(pud, address);
 349         if (pmd_none_or_clear_bad(pmd))
 350                 return 1;
 351         pte = pte_offset_kernel(pmd, address);
 352         entry = *pte;
 353         if (unlikely(pte_none(entry) || pte_not_present(entry)))
 354                 return 1;
 355         if (unlikely(writeaccess && !pte_write(entry)))
 356                 return 1;
 357
 358         if (writeaccess)
 359                 entry = pte_mkdirty(entry);
 360         entry = pte_mkyoung(entry);
 361
 362         set_pte(pte, entry);
 363
 364 #if defined(CONFIG_CPU_SH4) && !defined(CONFIG_SMP)
 365         /*
 366          * SH-4 does not set MMUCR.RC to the corresponding TLB entry in
 367          * the case of an initial page write exception, so we need to
 368          * flush it in order to avoid potential TLB entry duplication.
 369          */
 370         if (writeaccess == 2)
 371                 local_flush_tlb_one(get_asid(), address & PAGE_MASK);
 372 #endif
 373
 374         update_mmu_cache(NULL, address, entry);
 375
 376         return 0;
 377 }