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