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