arch/um/kernel/trap.c

   1 /*
   2  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include <linux/mm.h>
   7 #include <linux/sched.h>
   8 #include <linux/hardirq.h>
   9 #include <linux/module.h>
  10 #include <asm/current.h>
  11 #include <asm/pgtable.h>
  12 #include <asm/tlbflush.h>
  13 #include <arch.h>
  14 #include <as-layout.h>
  15 #include <kern_util.h>
  16 #include <os.h>
  17 #include <skas.h>
  18
  19 /*
  20  * Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by
  21  * segv().
  22  */
  23 int handle_page_fault(unsigned long address, unsigned long ip,
  24                       int is_write, int is_user, int *code_out)
  25 {
  26         struct mm_struct *mm = current->mm;
  27         struct vm_area_struct *vma;
  28         pgd_t *pgd;
  29         pud_t *pud;
  30         pmd_t *pmd;
  31         pte_t *pte;
  32         int err = -EFAULT;
  33         unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
  34
  35         *code_out = SEGV_MAPERR;
  36
  37         /*
  38          * If the fault was during atomic operation, don't take the fault, just
  39          * fail.
  40          */
  41         if (in_atomic())
  42                 goto out_nosemaphore;
  43
  44         if (is_user)
  45                 flags |= FAULT_FLAG_USER;
  46 retry:
  47         down_read(&mm->mmap_sem);
  48         vma = find_vma(mm, address);
  49         if (!vma)
  50                 goto out;
  51         else if (vma->vm_start <= address)
  52                 goto good_area;
  53         else if (!(vma->vm_flags & VM_GROWSDOWN))
  54                 goto out;
  55         else if (is_user && !ARCH_IS_STACKGROW(address))
  56                 goto out;
  57         else if (expand_stack(vma, address))
  58                 goto out;
  59
  60 good_area:
  61         *code_out = SEGV_ACCERR;
  62         if (is_write) {
  63                 if (!(vma->vm_flags & VM_WRITE))
  64                         goto out;
  65                 flags |= FAULT_FLAG_WRITE;
  66         } else {
  67                 /* Don't require VM_READ|VM_EXEC for write faults! */
  68                 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
  69                         goto out;
  70         }
  71
  72         do {
  73                 int fault;
  74
  75                 fault = handle_mm_fault(mm, vma, address, flags);
  76
  77                 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
  78                         goto out_nosemaphore;
  79
  80                 if (unlikely(fault & VM_FAULT_ERROR)) {
  81                         if (fault & VM_FAULT_OOM) {
  82                                 goto out_of_memory;
  83                         } else if (fault & VM_FAULT_SIGBUS) {
  84                                 err = -EACCES;
  85                                 goto out;
  86                         }
  87                         BUG();
  88                 }
  89                 if (flags & FAULT_FLAG_ALLOW_RETRY) {
  90                         if (fault & VM_FAULT_MAJOR)
  91                                 current->maj_flt++;
  92                         else
  93                                 current->min_flt++;
  94                         if (fault & VM_FAULT_RETRY) {
  95                                 flags &= ~FAULT_FLAG_ALLOW_RETRY;
  96                                 flags |= FAULT_FLAG_TRIED;
  97
  98                                 goto retry;
  99                         }
 100                 }
 101
 102                 pgd = pgd_offset(mm, address);
 103                 pud = pud_offset(pgd, address);
 104                 pmd = pmd_offset(pud, address);
 105                 pte = pte_offset_kernel(pmd, address);
 106         } while (!pte_present(*pte));
 107         err = 0;
 108         /*
 109          * The below warning was added in place of
 110          *      pte_mkyoung(); if (is_write) pte_mkdirty();
 111          * If it's triggered, we'd see normally a hang here (a clean pte is
 112          * marked read-only to emulate the dirty bit).
 113          * However, the generic code can mark a PTE writable but clean on a
 114          * concurrent read fault, triggering this harmlessly. So comment it out.
 115          */
 116 #if 0
 117         WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
 118 #endif
 119         flush_tlb_page(vma, address);
 120 out:
 121         up_read(&mm->mmap_sem);
 122 out_nosemaphore:
 123         return err;
 124
 125 out_of_memory:
 126         /*
 127          * We ran out of memory, call the OOM killer, and return the userspace
 128          * (which will retry the fault, or kill us if we got oom-killed).
 129          */
 130         up_read(&mm->mmap_sem);
 131         if (!is_user)
 132                 goto out_nosemaphore;
 133         pagefault_out_of_memory();
 134         return 0;
 135 }
 136 EXPORT_SYMBOL(handle_page_fault);
 137
 138 static void show_segv_info(struct uml_pt_regs *regs)
 139 {
 140         struct task_struct *tsk = current;
 141         struct faultinfo *fi = UPT_FAULTINFO(regs);
 142
 143         if (!unhandled_signal(tsk, SIGSEGV))
 144                 return;
 145
 146         if (!printk_ratelimit())
 147                 return;
 148
 149         printk("%s%s[%d]: segfault at %lx ip %p sp %p error %x",
 150                 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 151                 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
 152                 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
 153                 fi->error_code);
 154
 155         print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
 156         printk(KERN_CONT "\n");
 157 }
 158
 159 static void bad_segv(struct faultinfo fi, unsigned long ip)
 160 {
 161         struct siginfo si;
 162
 163         si.si_signo = SIGSEGV;
 164         si.si_code = SEGV_ACCERR;
 165         si.si_addr = (void __user *) FAULT_ADDRESS(fi);
 166         current->thread.arch.faultinfo = fi;
 167         force_sig_info(SIGSEGV, &si, current);
 168 }
 169
 170 void fatal_sigsegv(void)
 171 {
 172         force_sigsegv(SIGSEGV, current);
 173         do_signal();
 174         /*
 175          * This is to tell gcc that we're not returning - do_signal
 176          * can, in general, return, but in this case, it's not, since
 177          * we just got a fatal SIGSEGV queued.
 178          */
 179         os_dump_core();
 180 }
 181
 182 void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 183 {
 184         struct faultinfo * fi = UPT_FAULTINFO(regs);
 185
 186         if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
 187                 show_segv_info(regs);
 188                 bad_segv(*fi, UPT_IP(regs));
 189                 return;
 190         }
 191         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
 192 }
 193
 194 /*
 195  * We give a *copy* of the faultinfo in the regs to segv.
 196  * This must be done, since nesting SEGVs could overwrite
 197  * the info in the regs. A pointer to the info then would
 198  * give us bad data!
 199  */
 200 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
 201                    struct uml_pt_regs *regs)
 202 {
 203         struct siginfo si;
 204         jmp_buf *catcher;
 205         int err;
 206         int is_write = FAULT_WRITE(fi);
 207         unsigned long address = FAULT_ADDRESS(fi);
 208
 209         if (!is_user && (address >= start_vm) && (address < end_vm)) {
 210                 flush_tlb_kernel_vm();
 211                 return 0;
 212         }
 213         else if (current->mm == NULL) {
 214                 show_regs(container_of(regs, struct pt_regs, regs));
 215                 panic("Segfault with no mm");
 216         }
 217
 218         if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
 219                 err = handle_page_fault(address, ip, is_write, is_user,
 220                                         &si.si_code);
 221         else {
 222                 err = -EFAULT;
 223                 /*
 224                  * A thread accessed NULL, we get a fault, but CR2 is invalid.
 225                  * This code is used in __do_copy_from_user() of TT mode.
 226                  * XXX tt mode is gone, so maybe this isn't needed any more
 227                  */
 228                 address = 0;
 229         }
 230
 231         catcher = current->thread.fault_catcher;
 232         if (!err)
 233                 return 0;
 234         else if (catcher != NULL) {
 235                 current->thread.fault_addr = (void *) address;
 236                 UML_LONGJMP(catcher, 1);
 237         }
 238         else if (current->thread.fault_addr != NULL)
 239                 panic("fault_addr set but no fault catcher");
 240         else if (!is_user && arch_fixup(ip, regs))
 241                 return 0;
 242
 243         if (!is_user) {
 244                 show_regs(container_of(regs, struct pt_regs, regs));
 245                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 246                       address, ip);
 247         }
 248
 249         show_segv_info(regs);
 250
 251         if (err == -EACCES) {
 252                 si.si_signo = SIGBUS;
 253                 si.si_errno = 0;
 254                 si.si_code = BUS_ADRERR;
 255                 si.si_addr = (void __user *)address;
 256                 current->thread.arch.faultinfo = fi;
 257                 force_sig_info(SIGBUS, &si, current);
 258         } else {
 259                 BUG_ON(err != -EFAULT);
 260                 si.si_signo = SIGSEGV;
 261                 si.si_addr = (void __user *) address;
 262                 current->thread.arch.faultinfo = fi;
 263                 force_sig_info(SIGSEGV, &si, current);
 264         }
 265         return 0;
 266 }
 267
 268 void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
 269 {
 270         struct faultinfo *fi;
 271         struct siginfo clean_si;
 272
 273         if (!UPT_IS_USER(regs)) {
 274                 if (sig == SIGBUS)
 275                         printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
 276                                "mount likely just ran out of space\n");
 277                 panic("Kernel mode signal %d", sig);
 278         }
 279
 280         arch_examine_signal(sig, regs);
 281
 282         memset(&clean_si, 0, sizeof(clean_si));
 283         clean_si.si_signo = si->si_signo;
 284         clean_si.si_errno = si->si_errno;
 285         clean_si.si_code = si->si_code;
 286         switch (sig) {
 287         case SIGILL:
 288         case SIGFPE:
 289         case SIGSEGV:
 290         case SIGBUS:
 291         case SIGTRAP:
 292                 fi = UPT_FAULTINFO(regs);
 293                 clean_si.si_addr = (void __user *) FAULT_ADDRESS(*fi);
 294                 current->thread.arch.faultinfo = *fi;
 295 #ifdef __ARCH_SI_TRAPNO
 296                 clean_si.si_trapno = si->si_trapno;
 297 #endif
 298                 break;
 299         default:
 300                 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d)\n",
 301                         sig, si->si_code);
 302         }
 303
 304         force_sig_info(sig, &clean_si, current);
 305 }
 306
 307 void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
 308 {
 309         if (current->thread.fault_catcher != NULL)
 310                 UML_LONGJMP(current->thread.fault_catcher, 1);
 311         else
 312                 relay_signal(sig, si, regs);
 313 }
 314
 315 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
 316 {
 317         do_IRQ(WINCH_IRQ, regs);
 318 }
 319
 320 void trap_init(void)
 321 {
 322 }