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