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 "user_util.h"
  22 #include "kern_util.h"
  23 #include "kern.h"
  24 #include "chan_kern.h"
  25 #include "mconsole_kern.h"
  26 #include "2_5compat.h"
  27 #include "mem.h"
  28 #include "mem_kern.h"
  29
  30 int handle_page_fault(unsigned long address, unsigned long ip,
  31                       int is_write, int is_user, int *code_out)
  32 {
  33         struct mm_struct *mm = current->mm;
  34         struct vm_area_struct *vma;
  35         pgd_t *pgd;
  36         pud_t *pud;
  37         pmd_t *pmd;
  38         pte_t *pte;
  39         unsigned long page;
  40         int err = -EFAULT;
  41
  42         *code_out = SEGV_MAPERR;
  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         page = address & PAGE_MASK;
  61         pgd = pgd_offset(mm, page);
  62         pud = pud_offset(pgd, page);
  63         pmd = pmd_offset(pud, page);
  64         do {
  65  survive:
  66                 switch (handle_mm_fault(mm, vma, address, is_write)){
  67                 case VM_FAULT_MINOR:
  68                         current->min_flt++;
  69                         break;
  70                 case VM_FAULT_MAJOR:
  71                         current->maj_flt++;
  72                         break;
  73                 case VM_FAULT_SIGBUS:
  74                         err = -EACCES;
  75                         goto out;
  76                 case VM_FAULT_OOM:
  77                         err = -ENOMEM;
  78                         goto out_of_memory;
  79                 default:
  80                         BUG();
  81                 }
  82                 pgd = pgd_offset(mm, page);
  83                 pud = pud_offset(pgd, page);
  84                 pmd = pmd_offset(pud, page);
  85                 pte = pte_offset_kernel(pmd, page);
  86         } while(!pte_present(*pte));
  87         err = 0;
  88         *pte = pte_mkyoung(*pte);
  89         if(pte_write(*pte)) *pte = pte_mkdirty(*pte);
  90         flush_tlb_page(vma, page);
  91  out:
  92         up_read(&mm->mmap_sem);
  93         return(err);
  94
  95 /*
  96  * We ran out of memory, or some other thing happened to us that made
  97  * us unable to handle the page fault gracefully.
  98  */
  99 out_of_memory:
 100         if (current->pid == 1) {
 101                 up_read(&mm->mmap_sem);
 102                 yield();
 103                 down_read(&mm->mmap_sem);
 104                 goto survive;
 105         }
 106         goto out;
 107 }
 108
 109 LIST_HEAD(physmem_remappers);
 110
 111 void register_remapper(struct remapper *info)
 112 {
 113         list_add(&info->list, &physmem_remappers);
 114 }
 115
 116 static int check_remapped_addr(unsigned long address, int is_write)
 117 {
 118         struct remapper *remapper;
 119         struct list_head *ele;
 120         __u64 offset;
 121         int fd;
 122
 123         fd = phys_mapping(__pa(address), &offset);
 124         if(fd == -1)
 125                 return(0);
 126
 127         list_for_each(ele, &physmem_remappers){
 128                 remapper = list_entry(ele, struct remapper, list);
 129                 if((*remapper->proc)(fd, address, is_write, offset))
 130                         return(1);
 131         }
 132
 133         return(0);
 134 }
 135
 136 /*
 137  * We give a *copy* of the faultinfo in the regs to segv.
 138  * This must be done, since nesting SEGVs could overwrite
 139  * the info in the regs. A pointer to the info then would
 140  * give us bad data!
 141  */
 142 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user, void *sc)
 143 {
 144         struct siginfo si;
 145         void *catcher;
 146         int err;
 147         int is_write = FAULT_WRITE(fi);
 148         unsigned long address = FAULT_ADDRESS(fi);
 149
 150         if(!is_user && (address >= start_vm) && (address < end_vm)){
 151                 flush_tlb_kernel_vm();
 152                 return(0);
 153         }
 154         else if(check_remapped_addr(address & PAGE_MASK, is_write))
 155                 return(0);
 156         else if(current->mm == NULL)
 157                 panic("Segfault with no mm");
 158         err = handle_page_fault(address, ip, is_write, is_user, &si.si_code);
 159
 160         catcher = current->thread.fault_catcher;
 161         if(!err)
 162                 return(0);
 163         else if(catcher != NULL){
 164                 current->thread.fault_addr = (void *) address;
 165                 do_longjmp(catcher, 1);
 166         }
 167         else if(current->thread.fault_addr != NULL)
 168                 panic("fault_addr set but no fault catcher");
 169         else if(!is_user && arch_fixup(ip, sc))
 170                 return(0);
 171
 172         if(!is_user)
 173                 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
 174                       address, ip);
 175
 176         if(err == -EACCES){
 177                 si.si_signo = SIGBUS;
 178                 si.si_errno = 0;
 179                 si.si_code = BUS_ADRERR;
 180                 si.si_addr = (void *)address;
 181                 current->thread.arch.faultinfo = fi;
 182                 force_sig_info(SIGBUS, &si, current);
 183         }
 184         else if(err == -ENOMEM){
 185                 printk("VM: killing process %s\n", current->comm);
 186                 do_exit(SIGKILL);
 187         }
 188         else {
 189                 si.si_signo = SIGSEGV;
 190                 si.si_addr = (void *) address;
 191                 current->thread.arch.faultinfo = fi;
 192                 force_sig_info(SIGSEGV, &si, current);
 193         }
 194         return(0);
 195 }
 196
 197 void bad_segv(struct faultinfo fi, unsigned long ip)
 198 {
 199         struct siginfo si;
 200
 201         si.si_signo = SIGSEGV;
 202         si.si_code = SEGV_ACCERR;
 203         si.si_addr = (void *) FAULT_ADDRESS(fi);
 204         current->thread.arch.faultinfo = fi;
 205         force_sig_info(SIGSEGV, &si, current);
 206 }
 207
 208 void relay_signal(int sig, union uml_pt_regs *regs)
 209 {
 210         if(arch_handle_signal(sig, regs)) return;
 211         if(!UPT_IS_USER(regs))
 212                 panic("Kernel mode signal %d", sig);
 213         current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
 214         force_sig(sig, current);
 215 }
 216
 217 void bus_handler(int sig, union uml_pt_regs *regs)
 218 {
 219         if(current->thread.fault_catcher != NULL)
 220                 do_longjmp(current->thread.fault_catcher, 1);
 221         else relay_signal(sig, regs);
 222 }
 223
 224 void winch(int sig, union uml_pt_regs *regs)
 225 {
 226         do_IRQ(WINCH_IRQ, regs);
 227 }
 228
 229 void trap_init(void)
 230 {
 231 }
 232
 233 DEFINE_SPINLOCK(trap_lock);
 234
 235 static int trap_index = 0;
 236
 237 int next_trap_index(int limit)
 238 {
 239         int ret;
 240
 241         spin_lock(&trap_lock);
 242         ret = trap_index;
 243         if(++trap_index == limit)
 244                 trap_index = 0;
 245         spin_unlock(&trap_lock);
 246         return(ret);
 247 }
 248
 249 /*
 250  * Overrides for Emacs so that we follow Linus's tabbing style.
 251  * Emacs will notice this stuff at the end of the file and automatically
 252  * adjust the settings for this buffer only.  This must remain at the end
 253  * of the file.
 254  * ---------------------------------------------------------------------------
 255  * Local variables:
 256  * c-file-style: "linux"
 257  * End:
 258  */