arch/um/kernel/trap.c

   1 /*
   2  * Copyright (C) 2000, 2001 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 #include "linux/kernel.h"
   7 #include "asm/errno.h"
   8 #include "linux/sched.h"
   9 #include "linux/mm.h"
  10 #include "linux/spinlock.h"
  11 #include "linux/init.h"
  12 #include "linux/ptrace.h"
  13 #include "asm/semaphore.h"
  14 #include "asm/pgtable.h"
  15 #include "asm/pgalloc.h"
  16 #include "asm/tlbflush.h"
  17 #include "asm/a.out.h"
  18 #include "asm/current.h"
  19 #include "asm/irq.h"
  20 #include "sysdep/sigcontext.h"
  21 #include "kern_util.h"
  22 #include "as-layout.h"
  23 #include "arch.h"
  24 #include "kern.h"
  25 #include "chan_kern.h"
  26 #include "mconsole_kern.h"
  27 #include "mem.h"
  28 #include "mem_kern.h"
  29 #include "sysdep/sigcontext.h"
  30 #include "sysdep/ptrace.h"
  31 #include "os.h"
  32 #include "skas.h"
  33 #include "os.h"
  34
  35 /* Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by segv(). */
  36 int handle_page_fault(unsigned long address, unsigned long ip,
  37                       int is_write, int is_user, int *code_out)
  38 {
  39         struct mm_struct *mm = current->mm;
  40         struct vm_area_struct *vma;
  41         pgd_t *pgd;
  42         pud_t *pud;
  43         pmd_t *pmd;
  44         pte_t *pte;
  45         int err = -EFAULT;
  46
  47         *code_out = SEGV_MAPERR;
  48
  49         /* If the fault was during atomic operation, don't take the fault, just
  50          * fail. */
  51         if (in_atomic())
  52                 goto out_nosemaphore;
  53
  54         down_read(&mm->mmap_sem);
  55         vma = find_vma(mm, address);
  56         if(!vma)
  57                 goto out;
  58         else if(vma->vm_start <= address)
  59                 goto good_area;
  60         else if(!(vma->vm_flags & VM_GROWSDOWN))
  61                 goto out;
  62         else if(is_user && !ARCH_IS_STACKGROW(address))
  63                 goto out;
  64         else if(expand_stack(vma, address))
  65                 goto out;
  66
  67 good_area:
  68         *code_out = SEGV_ACCERR;
  69         if(is_write && !(vma->vm_flags & VM_WRITE))
  70                 goto out;
  71
  72         /* Don't require VM_READ|VM_EXEC for write faults! */
  73         if(!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
  74                 goto out;
  75
  76         do {
  77                 int fault;
  78 survive:
  79                 fault = handle_mm_fault(mm, vma, address, is_write);
  80                 if (unlikely(fault & VM_FAULT_ERROR)) {
  81                         if (fault & VM_FAULT_OOM) {
  82                                 err = -ENOMEM;
  83                                 goto out_of_memory;
  84                         } else if (fault & VM_FAULT_SIGBUS) {
  85                                 err = -EACCES;
  86                                 goto out;
  87                         }
  88                         BUG();
  89                 }
  90                 if (fault & VM_FAULT_MAJOR)
  91                         current->maj_flt++;
  92                 else
  93                         current->min_flt++;
  94
  95                 pgd = pgd_offset(mm, address);
  96                 pud = pud_offset(pgd, address);
  97                 pmd = pmd_offset(pud, address);
  98                 pte = pte_offset_kernel(pmd, address);
  99         } while(!pte_present(*pte));
 100         err = 0;
 101         /* The below warning was added in place of
 102          *      pte_mkyoung(); if (is_write) pte_mkdirty();
 103          * If it's triggered, we'd see normally a hang here (a clean pte is
 104          * marked read-only to emulate the dirty bit).
 105          * However, the generic code can mark a PTE writable but clean on a
 106          * concurrent read fault, triggering this harmlessly. So comment it out.
 107          */
 108 #if 0
 109         WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
 110 #endif
 111         flush_tlb_page(vma, address);
 112 out:
 113         up_read(&mm->mmap_sem);
 114 out_nosemaphore:
 115         return(err);
 116
 117 /*
 118  * We ran out of memory, or some other thing happened to us that made
 119  * us unable to handle the page fault gracefully.
 120  */
 121 out_of_memory:
 122         if (is_init(current)) {
 123                 up_read(&mm->mmap_sem);
 124                 yield();
 125                 down_read(&mm->mmap_sem);
 126                 goto survive;
 127         }
 128         goto out;
 129 }
 130
 131 static void bad_segv(struct faultinfo fi, unsigned long ip)
 132 {
 133         struct siginfo si;
 134
 135         si.si_signo = SIGSEGV;
 136         si.si_code = SEGV_ACCERR;
 137         si.si_addr = (void __user *) FAULT_ADDRESS(fi);
 138         current->thread.arch.faultinfo = fi;
 139         force_sig_info(SIGSEGV, &si, current);
 140 }
 141
 142 static void segv_handler(int sig, union uml_pt_regs *regs)
 143 {
 144         struct faultinfo * fi = UPT_FAULTINFO(regs);
 145
 146         if(UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)){
 147                 bad_segv(*fi, UPT_IP(regs));
 148                 return;
 149         }
 150         segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
 151 }
 152
 153 /*
 154  * We give a *copy* of the faultinfo in the regs to segv.
 155  * This must be done, since nesting SEGVs could overwrite
 156  * the info in the regs. A pointer to the info then would
 157  * give us bad data!
 158  */
 159 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
 160                    union uml_pt_regs *regs)
 161 {
 162         struct siginfo si;
 163         void *catcher;
 164         int err;
 165         int is_write = FAULT_WRITE(fi);
 166         unsigned long address = FAULT_ADDRESS(fi);
 167
 168         if(!is_user && (address >= start_vm) && (address < end_vm)){
 169                 flush_tlb_kernel_vm();
 170                 return 0;
 171         }
 172         else if(current->mm == NULL) {
 173                 show_regs(container_of(regs, struct pt_regs, regs));
 174                 panic("Segfault with no mm");
 175         }
 176
 177         if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
 178                 err = handle_page_fault(address, ip, is_write, is_user, &si.si_code);
 179         else {
 180                 err = -EFAULT;
 181                 /* A thread accessed NULL, we get a fault, but CR2 is invalid.
 182                  * This code is used in __do_copy_from_user() of TT mode. */
 183                 address = 0;
 184         }
 185
 186         catcher = current->thread.fault_catcher;
 187         if(!err)
 188                 return 0;
 189         else if(catcher != NULL){
 190                 current->thread.fault_addr = (void *) address;
 191                 do_longjmp(catcher, 1);
 192         }
 193         else if(current->thread.fault_addr != NULL)
 194                 panic("fault_addr set but no fault catcher");
 195         else if(!is_user && arch_fixup(ip, regs))
 196                 return 0;
 197
 198         if(!is_user) {
 199                 show_regs(container_of(regs, struct pt_regs, regs));
 200                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 201                       address, ip);
 202         }
 203
 204         if (err == -EACCES) {
 205                 si.si_signo = SIGBUS;
 206                 si.si_errno = 0;
 207                 si.si_code = BUS_ADRERR;
 208                 si.si_addr = (void __user *)address;
 209                 current->thread.arch.faultinfo = fi;
 210                 force_sig_info(SIGBUS, &si, current);
 211         } else if (err == -ENOMEM) {
 212                 printk("VM: killing process %s\n", current->comm);
 213                 do_exit(SIGKILL);
 214         } else {
 215                 BUG_ON(err != -EFAULT);
 216                 si.si_signo = SIGSEGV;
 217                 si.si_addr = (void __user *) address;
 218                 current->thread.arch.faultinfo = fi;
 219                 force_sig_info(SIGSEGV, &si, current);
 220         }
 221         return 0;
 222 }
 223
 224 void relay_signal(int sig, union uml_pt_regs *regs)
 225 {
 226         if(arch_handle_signal(sig, regs))
 227                 return;
 228
 229         if(!UPT_IS_USER(regs)){
 230                 if(sig == SIGBUS)
 231                         printk("Bus error - the host /dev/shm or /tmp mount "
 232                                "likely just ran out of space\n");
 233                 panic("Kernel mode signal %d", sig);
 234         }
 235
 236         current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
 237         force_sig(sig, current);
 238 }
 239
 240 static void bus_handler(int sig, union uml_pt_regs *regs)
 241 {
 242         if(current->thread.fault_catcher != NULL)
 243                 do_longjmp(current->thread.fault_catcher, 1);
 244         else relay_signal(sig, regs);
 245 }
 246
 247 static void winch(int sig, union uml_pt_regs *regs)
 248 {
 249         do_IRQ(WINCH_IRQ, regs);
 250 }
 251
 252 const struct kern_handlers handlinfo_kern = {
 253         .relay_signal = relay_signal,
 254         .winch = winch,
 255         .bus_handler = bus_handler,
 256         .page_fault = segv_handler,
 257         .sigio_handler = sigio_handler,
 258         .timer_handler = timer_handler
 259 };
 260
 261 void trap_init(void)
 262 {
 263 }