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