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 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 } 115 116 static void bad_segv(struct faultinfo fi, unsigned long ip) 117 { 118 struct siginfo si; 119 120 si.si_signo = SIGSEGV; 121 si.si_code = SEGV_ACCERR; 122 si.si_addr = (void __user *) FAULT_ADDRESS(fi); 123 current->thread.arch.faultinfo = fi; 124 force_sig_info(SIGSEGV, &si, current); 125 } 126 127 void fatal_sigsegv(void) 128 { 129 force_sigsegv(SIGSEGV, current); 130 do_signal(); 131 /* 132 * This is to tell gcc that we're not returning - do_signal 133 * can, in general, return, but in this case, it's not, since 134 * we just got a fatal SIGSEGV queued. 135 */ 136 os_dump_core(); 137 } 138 139 void segv_handler(int sig, struct uml_pt_regs *regs) 140 { 141 struct faultinfo * fi = UPT_FAULTINFO(regs); 142 143 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) { 144 bad_segv(*fi, UPT_IP(regs)); 145 return; 146 } 147 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs); 148 } 149 150 /* 151 * We give a *copy* of the faultinfo in the regs to segv. 152 * This must be done, since nesting SEGVs could overwrite 153 * the info in the regs. A pointer to the info then would 154 * give us bad data! 155 */ 156 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user, 157 struct uml_pt_regs *regs) 158 { 159 struct siginfo si; 160 jmp_buf *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 show_regs(container_of(regs, struct pt_regs, regs)); 171 panic("Segfault with no mm"); 172 } 173 174 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi)) 175 err = handle_page_fault(address, ip, is_write, is_user, 176 &si.si_code); 177 else { 178 err = -EFAULT; 179 /* 180 * A thread accessed NULL, we get a fault, but CR2 is invalid. 181 * This code is used in __do_copy_from_user() of TT mode. 182 * XXX tt mode is gone, so maybe this isn't needed any more 183 */ 184 address = 0; 185 } 186 187 catcher = current->thread.fault_catcher; 188 if (!err) 189 return 0; 190 else if (catcher != NULL) { 191 current->thread.fault_addr = (void *) address; 192 UML_LONGJMP(catcher, 1); 193 } 194 else if (current->thread.fault_addr != NULL) 195 panic("fault_addr set but no fault catcher"); 196 else if (!is_user && arch_fixup(ip, regs)) 197 return 0; 198 199 if (!is_user) { 200 show_regs(container_of(regs, struct pt_regs, regs)); 201 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx", 202 address, ip); 203 } 204 205 if (err == -EACCES) { 206 si.si_signo = SIGBUS; 207 si.si_errno = 0; 208 si.si_code = BUS_ADRERR; 209 si.si_addr = (void __user *)address; 210 current->thread.arch.faultinfo = fi; 211 force_sig_info(SIGBUS, &si, current); 212 } else { 213 BUG_ON(err != -EFAULT); 214 si.si_signo = SIGSEGV; 215 si.si_addr = (void __user *) address; 216 current->thread.arch.faultinfo = fi; 217 force_sig_info(SIGSEGV, &si, current); 218 } 219 return 0; 220 } 221 222 void relay_signal(int sig, struct uml_pt_regs *regs) 223 { 224 if (!UPT_IS_USER(regs)) { 225 if (sig == SIGBUS) 226 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp " 227 "mount likely just ran out of space\n"); 228 panic("Kernel mode signal %d", sig); 229 } 230 231 arch_examine_signal(sig, regs); 232 233 current->thread.arch.faultinfo = *UPT_FAULTINFO(regs); 234 force_sig(sig, current); 235 } 236 237 void bus_handler(int sig, struct uml_pt_regs *regs) 238 { 239 if (current->thread.fault_catcher != NULL) 240 UML_LONGJMP(current->thread.fault_catcher, 1); 241 else relay_signal(sig, regs); 242 } 243 244 void winch(int sig, struct uml_pt_regs *regs) 245 { 246 do_IRQ(WINCH_IRQ, regs); 247 } 248 249 void trap_init(void) 250 { 251 } 252