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 survive:
  68                 fault = handle_mm_fault(mm, vma, address, is_write);
  69                 if (unlikely(fault & VM_FAULT_ERROR)) {
  70                         if (fault & VM_FAULT_OOM) {
  71                                 err = -ENOMEM;
  72                                 goto out_of_memory;
  73                         } else if (fault & VM_FAULT_SIGBUS) {
  74                                 err = -EACCES;
  75                                 goto out;
  76                         }
  77                         BUG();
  78                 }
  79                 if (fault & VM_FAULT_MAJOR)
  80                         current->maj_flt++;
  81                 else
  82                         current->min_flt++;
  83
  84                 pgd = pgd_offset(mm, address);
  85                 pud = pud_offset(pgd, address);
  86                 pmd = pmd_offset(pud, address);
  87                 pte = pte_offset_kernel(pmd, address);
  88         } while (!pte_present(*pte));
  89         err = 0;
  90         /*
  91          * The below warning was added in place of
  92          *      pte_mkyoung(); if (is_write) pte_mkdirty();
  93          * If it's triggered, we'd see normally a hang here (a clean pte is
  94          * marked read-only to emulate the dirty bit).
  95          * However, the generic code can mark a PTE writable but clean on a
  96          * concurrent read fault, triggering this harmlessly. So comment it out.
  97          */
  98 #if 0
  99         WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
 100 #endif
 101         flush_tlb_page(vma, address);
 102 out:
 103         up_read(&mm->mmap_sem);
 104 out_nosemaphore:
 105         return err;
 106
 107 /*
 108  * We ran out of memory, or some other thing happened to us that made
 109  * us unable to handle the page fault gracefully.
 110  */
 111 out_of_memory:
 112         if (is_global_init(current)) {
 113                 up_read(&mm->mmap_sem);
 114                 yield();
 115                 down_read(&mm->mmap_sem);
 116                 goto survive;
 117         }
 118         goto out;
 119 }
 120
 121 static void bad_segv(struct faultinfo fi, unsigned long ip)
 122 {
 123         struct siginfo si;
 124
 125         si.si_signo = SIGSEGV;
 126         si.si_code = SEGV_ACCERR;
 127         si.si_addr = (void __user *) FAULT_ADDRESS(fi);
 128         current->thread.arch.faultinfo = fi;
 129         force_sig_info(SIGSEGV, &si, current);
 130 }
 131
 132 void fatal_sigsegv(void)
 133 {
 134         force_sigsegv(SIGSEGV, current);
 135         do_signal();
 136         /*
 137          * This is to tell gcc that we're not returning - do_signal
 138          * can, in general, return, but in this case, it's not, since
 139          * we just got a fatal SIGSEGV queued.
 140          */
 141         os_dump_core();
 142 }
 143
 144 void segv_handler(int sig, struct uml_pt_regs *regs)
 145 {
 146         struct faultinfo * fi = UPT_FAULTINFO(regs);
 147
 148         if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
 149                 bad_segv(*fi, UPT_IP(regs));
 150                 return;
 151         }
 152         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
 153 }
 154
 155 /*
 156  * We give a *copy* of the faultinfo in the regs to segv.
 157  * This must be done, since nesting SEGVs could overwrite
 158  * the info in the regs. A pointer to the info then would
 159  * give us bad data!
 160  */
 161 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
 162                    struct uml_pt_regs *regs)
 163 {
 164         struct siginfo si;
 165         jmp_buf *catcher;
 166         int err;
 167         int is_write = FAULT_WRITE(fi);
 168         unsigned long address = FAULT_ADDRESS(fi);
 169
 170         if (!is_user && (address >= start_vm) && (address < end_vm)) {
 171                 flush_tlb_kernel_vm();
 172                 return 0;
 173         }
 174         else if (current->mm == NULL) {
 175                 show_regs(container_of(regs, struct pt_regs, regs));
 176                 panic("Segfault with no mm");
 177         }
 178
 179         if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
 180                 err = handle_page_fault(address, ip, is_write, is_user,
 181                                         &si.si_code);
 182         else {
 183                 err = -EFAULT;
 184                 /*
 185                  * A thread accessed NULL, we get a fault, but CR2 is invalid.
 186                  * This code is used in __do_copy_from_user() of TT mode.
 187                  * XXX tt mode is gone, so maybe this isn't needed any more
 188                  */
 189                 address = 0;
 190         }
 191
 192         catcher = current->thread.fault_catcher;
 193         if (!err)
 194                 return 0;
 195         else if (catcher != NULL) {
 196                 current->thread.fault_addr = (void *) address;
 197                 UML_LONGJMP(catcher, 1);
 198         }
 199         else if (current->thread.fault_addr != NULL)
 200                 panic("fault_addr set but no fault catcher");
 201         else if (!is_user && arch_fixup(ip, regs))
 202                 return 0;
 203
 204         if (!is_user) {
 205                 show_regs(container_of(regs, struct pt_regs, regs));
 206                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 207                       address, ip);
 208         }
 209
 210         if (err == -EACCES) {
 211                 si.si_signo = SIGBUS;
 212                 si.si_errno = 0;
 213                 si.si_code = BUS_ADRERR;
 214                 si.si_addr = (void __user *)address;
 215                 current->thread.arch.faultinfo = fi;
 216                 force_sig_info(SIGBUS, &si, current);
 217         } else if (err == -ENOMEM) {
 218                 printk(KERN_INFO "VM: killing process %s\n", current->comm);
 219                 do_exit(SIGKILL);
 220         } else {
 221                 BUG_ON(err != -EFAULT);
 222                 si.si_signo = SIGSEGV;
 223                 si.si_addr = (void __user *) address;
 224                 current->thread.arch.faultinfo = fi;
 225                 force_sig_info(SIGSEGV, &si, current);
 226         }
 227         return 0;
 228 }
 229
 230 void relay_signal(int sig, struct uml_pt_regs *regs)
 231 {
 232         if (!UPT_IS_USER(regs)) {
 233                 if (sig == SIGBUS)
 234                         printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
 235                                "mount likely just ran out of space\n");
 236                 panic("Kernel mode signal %d", sig);
 237         }
 238
 239         arch_examine_signal(sig, regs);
 240
 241         current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
 242         force_sig(sig, current);
 243 }
 244
 245 void bus_handler(int sig, struct uml_pt_regs *regs)
 246 {
 247         if (current->thread.fault_catcher != NULL)
 248                 UML_LONGJMP(current->thread.fault_catcher, 1);
 249         else relay_signal(sig, regs);
 250 }
 251
 252 void winch(int sig, struct uml_pt_regs *regs)
 253 {
 254         do_IRQ(WINCH_IRQ, regs);
 255 }
 256
 257 void trap_init(void)
 258 {
 259 }