2 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
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>
13 #include "as-layout.h"
14 #include "kern_util.h"
16 #include "sysdep/sigcontext.h"
19 * Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by
22 int handle_page_fault(unsigned long address, unsigned long ip,
23 int is_write, int is_user, int *code_out)
25 struct mm_struct *mm = current->mm;
26 struct vm_area_struct *vma;
33 *code_out = SEGV_MAPERR;
36 * If the fault was during atomic operation, don't take the fault, just
42 down_read(&mm->mmap_sem);
43 vma = find_vma(mm, address);
46 else if (vma->vm_start <= address)
48 else if (!(vma->vm_flags & VM_GROWSDOWN))
50 else if (is_user && !ARCH_IS_STACKGROW(address))
52 else if (expand_stack(vma, address))
56 *code_out = SEGV_ACCERR;
57 if (is_write && !(vma->vm_flags & VM_WRITE))
60 /* Don't require VM_READ|VM_EXEC for write faults! */
61 if (!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
67 fault = handle_mm_fault(mm, vma, address, is_write);
68 if (unlikely(fault & VM_FAULT_ERROR)) {
69 if (fault & VM_FAULT_OOM) {
72 } else if (fault & VM_FAULT_SIGBUS) {
78 if (fault & VM_FAULT_MAJOR)
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));
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.
98 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
100 flush_tlb_page(vma, address);
102 up_read(&mm->mmap_sem);
107 * We ran out of memory, or some other thing happened to us that made
108 * us unable to handle the page fault gracefully.
111 if (is_init(current)) {
112 up_read(&mm->mmap_sem);
114 down_read(&mm->mmap_sem);
120 static void bad_segv(struct faultinfo fi, unsigned long ip)
124 si.si_signo = SIGSEGV;
125 si.si_code = SEGV_ACCERR;
126 si.si_addr = (void __user *) FAULT_ADDRESS(fi);
127 current->thread.arch.faultinfo = fi;
128 force_sig_info(SIGSEGV, &si, current);
131 static void segv_handler(int sig, struct uml_pt_regs *regs)
133 struct faultinfo * fi = UPT_FAULTINFO(regs);
135 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
136 bad_segv(*fi, UPT_IP(regs));
139 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
143 * We give a *copy* of the faultinfo in the regs to segv.
144 * This must be done, since nesting SEGVs could overwrite
145 * the info in the regs. A pointer to the info then would
148 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
149 struct uml_pt_regs *regs)
154 int is_write = FAULT_WRITE(fi);
155 unsigned long address = FAULT_ADDRESS(fi);
157 if (!is_user && (address >= start_vm) && (address < end_vm)) {
158 flush_tlb_kernel_vm();
161 else if (current->mm == NULL) {
162 show_regs(container_of(regs, struct pt_regs, regs));
163 panic("Segfault with no mm");
166 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
167 err = handle_page_fault(address, ip, is_write, is_user,
172 * A thread accessed NULL, we get a fault, but CR2 is invalid.
173 * This code is used in __do_copy_from_user() of TT mode.
174 * XXX tt mode is gone, so maybe this isn't needed any more
179 catcher = current->thread.fault_catcher;
182 else if (catcher != NULL) {
183 current->thread.fault_addr = (void *) address;
184 UML_LONGJMP(catcher, 1);
186 else if (current->thread.fault_addr != NULL)
187 panic("fault_addr set but no fault catcher");
188 else if (!is_user && arch_fixup(ip, regs))
192 show_regs(container_of(regs, struct pt_regs, regs));
193 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
197 if (err == -EACCES) {
198 si.si_signo = SIGBUS;
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(KERN_INFO "VM: killing process %s\n", current->comm);
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);
217 void relay_signal(int sig, struct uml_pt_regs *regs)
219 if (arch_handle_signal(sig, regs))
222 if (!UPT_IS_USER(regs)) {
224 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
225 "mount likely just ran out of space\n");
226 panic("Kernel mode signal %d", sig);
229 current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
230 force_sig(sig, current);
233 static void bus_handler(int sig, struct uml_pt_regs *regs)
235 if (current->thread.fault_catcher != NULL)
236 UML_LONGJMP(current->thread.fault_catcher, 1);
237 else relay_signal(sig, regs);
240 static void winch(int sig, struct uml_pt_regs *regs)
242 do_IRQ(WINCH_IRQ, regs);
245 const struct kern_handlers handlinfo_kern = {
246 .relay_signal = relay_signal,
248 .bus_handler = bus_handler,
249 .page_fault = segv_handler,
250 .sigio_handler = sigio_handler,
251 .timer_handler = timer_handler