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"
17 #include "sysdep/sigcontext.h"
20 * Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by
23 int handle_page_fault(unsigned long address, unsigned long ip,
24 int is_write, int is_user, int *code_out)
26 struct mm_struct *mm = current->mm;
27 struct vm_area_struct *vma;
34 *code_out = SEGV_MAPERR;
37 * If the fault was during atomic operation, don't take the fault, just
43 down_read(&mm->mmap_sem);
44 vma = find_vma(mm, address);
47 else if (vma->vm_start <= address)
49 else if (!(vma->vm_flags & VM_GROWSDOWN))
51 else if (is_user && !ARCH_IS_STACKGROW(address))
53 else if (expand_stack(vma, address))
57 *code_out = SEGV_ACCERR;
58 if (is_write && !(vma->vm_flags & VM_WRITE))
61 /* Don't require VM_READ|VM_EXEC for write faults! */
62 if (!is_write && !(vma->vm_flags & (VM_READ | VM_EXEC)))
68 fault = handle_mm_fault(mm, vma, address, is_write);
69 if (unlikely(fault & VM_FAULT_ERROR)) {
70 if (fault & VM_FAULT_OOM) {
73 } else if (fault & VM_FAULT_SIGBUS) {
79 if (fault & VM_FAULT_MAJOR)
84 pgd = pgd_offset(mm, address);
85 pud = pud_offset(pgd, address);
86 pmd = pmd_offset(pud, address);
87 pte = pte_offset_kernel(pmd, address);
88 } while (!pte_present(*pte));
91 * The below warning was added in place of
92 * pte_mkyoung(); if (is_write) pte_mkdirty();
93 * If it's triggered, we'd see normally a hang here (a clean pte is
94 * marked read-only to emulate the dirty bit).
95 * However, the generic code can mark a PTE writable but clean on a
96 * concurrent read fault, triggering this harmlessly. So comment it out.
99 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
101 flush_tlb_page(vma, address);
103 up_read(&mm->mmap_sem);
108 * We ran out of memory, or some other thing happened to us that made
109 * us unable to handle the page fault gracefully.
112 if (is_global_init(current)) {
113 up_read(&mm->mmap_sem);
115 down_read(&mm->mmap_sem);
121 static void bad_segv(struct faultinfo fi, unsigned long ip)
125 si.si_signo = SIGSEGV;
126 si.si_code = SEGV_ACCERR;
127 si.si_addr = (void __user *) FAULT_ADDRESS(fi);
128 current->thread.arch.faultinfo = fi;
129 force_sig_info(SIGSEGV, &si, current);
132 void fatal_sigsegv(void)
134 force_sigsegv(SIGSEGV, current);
137 * This is to tell gcc that we're not returning - do_signal
138 * can, in general, return, but in this case, it's not, since
139 * we just got a fatal SIGSEGV queued.
144 void segv_handler(int sig, struct uml_pt_regs *regs)
146 struct faultinfo * fi = UPT_FAULTINFO(regs);
148 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
149 bad_segv(*fi, UPT_IP(regs));
152 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
156 * We give a *copy* of the faultinfo in the regs to segv.
157 * This must be done, since nesting SEGVs could overwrite
158 * the info in the regs. A pointer to the info then would
161 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
162 struct uml_pt_regs *regs)
167 int is_write = FAULT_WRITE(fi);
168 unsigned long address = FAULT_ADDRESS(fi);
170 if (!is_user && (address >= start_vm) && (address < end_vm)) {
171 flush_tlb_kernel_vm();
174 else if (current->mm == NULL) {
175 show_regs(container_of(regs, struct pt_regs, regs));
176 panic("Segfault with no mm");
179 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
180 err = handle_page_fault(address, ip, is_write, is_user,
185 * A thread accessed NULL, we get a fault, but CR2 is invalid.
186 * This code is used in __do_copy_from_user() of TT mode.
187 * XXX tt mode is gone, so maybe this isn't needed any more
192 catcher = current->thread.fault_catcher;
195 else if (catcher != NULL) {
196 current->thread.fault_addr = (void *) address;
197 UML_LONGJMP(catcher, 1);
199 else if (current->thread.fault_addr != NULL)
200 panic("fault_addr set but no fault catcher");
201 else if (!is_user && arch_fixup(ip, regs))
205 show_regs(container_of(regs, struct pt_regs, regs));
206 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
210 if (err == -EACCES) {
211 si.si_signo = SIGBUS;
213 si.si_code = BUS_ADRERR;
214 si.si_addr = (void __user *)address;
215 current->thread.arch.faultinfo = fi;
216 force_sig_info(SIGBUS, &si, current);
217 } else if (err == -ENOMEM) {
218 printk(KERN_INFO "VM: killing process %s\n", current->comm);
221 BUG_ON(err != -EFAULT);
222 si.si_signo = SIGSEGV;
223 si.si_addr = (void __user *) address;
224 current->thread.arch.faultinfo = fi;
225 force_sig_info(SIGSEGV, &si, current);
230 void relay_signal(int sig, struct uml_pt_regs *regs)
232 if (!UPT_IS_USER(regs)) {
234 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
235 "mount likely just ran out of space\n");
236 panic("Kernel mode signal %d", sig);
239 arch_examine_signal(sig, regs);
241 current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
242 force_sig(sig, current);
245 void bus_handler(int sig, struct uml_pt_regs *regs)
247 if (current->thread.fault_catcher != NULL)
248 UML_LONGJMP(current->thread.fault_catcher, 1);
249 else relay_signal(sig, regs);
252 void winch(int sig, struct uml_pt_regs *regs)
254 do_IRQ(WINCH_IRQ, regs);