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
  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 EXPORT_SYMBOL(handle_page_fault);
 116
 117 static void show_segv_info(struct uml_pt_regs *regs)
 118 {
 119         struct task_struct *tsk = current;
 120         struct faultinfo *fi = UPT_FAULTINFO(regs);
 121
 122         if (!unhandled_signal(tsk, SIGSEGV))
 123                 return;
 124
 125         if (!printk_ratelimit())
 126                 return;
 127
 128         printk("%s%s[%d]: segfault at %lx ip %p sp %p error %x",
 129                 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
 130                 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
 131                 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
 132                 fi->error_code);
 133
 134         print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
 135         printk(KERN_CONT "\n");
 136 }
 137
 138 static void bad_segv(struct faultinfo fi, unsigned long ip)
 139 {
 140         struct siginfo si;
 141
 142         si.si_signo = SIGSEGV;
 143         si.si_code = SEGV_ACCERR;
 144         si.si_addr = (void __user *) FAULT_ADDRESS(fi);
 145         current->thread.arch.faultinfo = fi;
 146         force_sig_info(SIGSEGV, &si, current);
 147 }
 148
 149 void fatal_sigsegv(void)
 150 {
 151         force_sigsegv(SIGSEGV, current);
 152         do_signal();
 153         /*
 154          * This is to tell gcc that we're not returning - do_signal
 155          * can, in general, return, but in this case, it's not, since
 156          * we just got a fatal SIGSEGV queued.
 157          */
 158         os_dump_core();
 159 }
 160
 161 void segv_handler(int sig, struct uml_pt_regs *regs)
 162 {
 163         struct faultinfo * fi = UPT_FAULTINFO(regs);
 164
 165         if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
 166                 show_segv_info(regs);
 167                 bad_segv(*fi, UPT_IP(regs));
 168                 return;
 169         }
 170         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
 171 }
 172
 173 /*
 174  * We give a *copy* of the faultinfo in the regs to segv.
 175  * This must be done, since nesting SEGVs could overwrite
 176  * the info in the regs. A pointer to the info then would
 177  * give us bad data!
 178  */
 179 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
 180                    struct uml_pt_regs *regs)
 181 {
 182         struct siginfo si;
 183         jmp_buf *catcher;
 184         int err;
 185         int is_write = FAULT_WRITE(fi);
 186         unsigned long address = FAULT_ADDRESS(fi);
 187
 188         if (!is_user && (address >= start_vm) && (address < end_vm)) {
 189                 flush_tlb_kernel_vm();
 190                 return 0;
 191         }
 192         else if (current->mm == NULL) {
 193                 show_regs(container_of(regs, struct pt_regs, regs));
 194                 panic("Segfault with no mm");
 195         }
 196
 197         if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
 198                 err = handle_page_fault(address, ip, is_write, is_user,
 199                                         &si.si_code);
 200         else {
 201                 err = -EFAULT;
 202                 /*
 203                  * A thread accessed NULL, we get a fault, but CR2 is invalid.
 204                  * This code is used in __do_copy_from_user() of TT mode.
 205                  * XXX tt mode is gone, so maybe this isn't needed any more
 206                  */
 207                 address = 0;
 208         }
 209
 210         catcher = current->thread.fault_catcher;
 211         if (!err)
 212                 return 0;
 213         else if (catcher != NULL) {
 214                 current->thread.fault_addr = (void *) address;
 215                 UML_LONGJMP(catcher, 1);
 216         }
 217         else if (current->thread.fault_addr != NULL)
 218                 panic("fault_addr set but no fault catcher");
 219         else if (!is_user && arch_fixup(ip, regs))
 220                 return 0;
 221
 222         if (!is_user) {
 223                 show_regs(container_of(regs, struct pt_regs, regs));
 224                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 225                       address, ip);
 226         }
 227
 228         show_segv_info(regs);
 229
 230         if (err == -EACCES) {
 231                 si.si_signo = SIGBUS;
 232                 si.si_errno = 0;
 233                 si.si_code = BUS_ADRERR;
 234                 si.si_addr = (void __user *)address;
 235                 current->thread.arch.faultinfo = fi;
 236                 force_sig_info(SIGBUS, &si, current);
 237         } else {
 238                 BUG_ON(err != -EFAULT);
 239                 si.si_signo = SIGSEGV;
 240                 si.si_addr = (void __user *) address;
 241                 current->thread.arch.faultinfo = fi;
 242                 force_sig_info(SIGSEGV, &si, current);
 243         }
 244         return 0;
 245 }
 246
 247 void relay_signal(int sig, struct uml_pt_regs *regs)
 248 {
 249         if (!UPT_IS_USER(regs)) {
 250                 if (sig == SIGBUS)
 251                         printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
 252                                "mount likely just ran out of space\n");
 253                 panic("Kernel mode signal %d", sig);
 254         }
 255
 256         arch_examine_signal(sig, regs);
 257
 258         current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
 259         force_sig(sig, current);
 260 }
 261
 262 void bus_handler(int sig, struct uml_pt_regs *regs)
 263 {
 264         if (current->thread.fault_catcher != NULL)
 265                 UML_LONGJMP(current->thread.fault_catcher, 1);
 266         else relay_signal(sig, regs);
 267 }
 268
 269 void winch(int sig, struct uml_pt_regs *regs)
 270 {
 271         do_IRQ(WINCH_IRQ, regs);
 272 }
 273
 274 void trap_init(void)
 275 {
 276 }