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 <linux/module.h>
10 #include <asm/current.h>
11 #include <asm/pgtable.h>
12 #include <asm/tlbflush.h>
14 #include "as-layout.h"
15 #include "kern_util.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 ? FAULT_FLAG_WRITE : 0);
69 if (unlikely(fault & VM_FAULT_ERROR)) {
70 if (fault & VM_FAULT_OOM) {
72 } else if (fault & VM_FAULT_SIGSEGV) {
74 } else if (fault & VM_FAULT_SIGBUS) {
80 if (fault & VM_FAULT_MAJOR)
85 pgd = pgd_offset(mm, address);
86 pud = pud_offset(pgd, address);
87 pmd = pmd_offset(pud, address);
88 pte = pte_offset_kernel(pmd, address);
89 } while (!pte_present(*pte));
92 * The below warning was added in place of
93 * pte_mkyoung(); if (is_write) pte_mkdirty();
94 * If it's triggered, we'd see normally a hang here (a clean pte is
95 * marked read-only to emulate the dirty bit).
96 * However, the generic code can mark a PTE writable but clean on a
97 * concurrent read fault, triggering this harmlessly. So comment it out.
100 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
102 flush_tlb_page(vma, address);
104 up_read(&mm->mmap_sem);
110 * We ran out of memory, call the OOM killer, and return the userspace
111 * (which will retry the fault, or kill us if we got oom-killed).
113 up_read(&mm->mmap_sem);
114 pagefault_out_of_memory();
117 EXPORT_SYMBOL(handle_page_fault);
119 static void show_segv_info(struct uml_pt_regs *regs)
121 struct task_struct *tsk = current;
122 struct faultinfo *fi = UPT_FAULTINFO(regs);
124 if (!unhandled_signal(tsk, SIGSEGV))
127 if (!printk_ratelimit())
130 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %x",
131 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
132 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
133 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
136 print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
137 printk(KERN_CONT "\n");
140 static void bad_segv(struct faultinfo fi, unsigned long ip)
144 si.si_signo = SIGSEGV;
145 si.si_code = SEGV_ACCERR;
146 si.si_addr = (void __user *) FAULT_ADDRESS(fi);
147 current->thread.arch.faultinfo = fi;
148 force_sig_info(SIGSEGV, &si, current);
151 void fatal_sigsegv(void)
153 force_sigsegv(SIGSEGV, current);
156 * This is to tell gcc that we're not returning - do_signal
157 * can, in general, return, but in this case, it's not, since
158 * we just got a fatal SIGSEGV queued.
163 void segv_handler(int sig, struct uml_pt_regs *regs)
165 struct faultinfo * fi = UPT_FAULTINFO(regs);
167 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
168 show_segv_info(regs);
169 bad_segv(*fi, UPT_IP(regs));
172 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
176 * We give a *copy* of the faultinfo in the regs to segv.
177 * This must be done, since nesting SEGVs could overwrite
178 * the info in the regs. A pointer to the info then would
181 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
182 struct uml_pt_regs *regs)
187 int is_write = FAULT_WRITE(fi);
188 unsigned long address = FAULT_ADDRESS(fi);
190 if (!is_user && (address >= start_vm) && (address < end_vm)) {
191 flush_tlb_kernel_vm();
194 else if (current->mm == NULL) {
195 show_regs(container_of(regs, struct pt_regs, regs));
196 panic("Segfault with no mm");
199 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
200 err = handle_page_fault(address, ip, is_write, is_user,
205 * A thread accessed NULL, we get a fault, but CR2 is invalid.
206 * This code is used in __do_copy_from_user() of TT mode.
207 * XXX tt mode is gone, so maybe this isn't needed any more
212 catcher = current->thread.fault_catcher;
215 else if (catcher != NULL) {
216 current->thread.fault_addr = (void *) address;
217 UML_LONGJMP(catcher, 1);
219 else if (current->thread.fault_addr != NULL)
220 panic("fault_addr set but no fault catcher");
221 else if (!is_user && arch_fixup(ip, regs))
225 show_regs(container_of(regs, struct pt_regs, regs));
226 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
230 show_segv_info(regs);
232 if (err == -EACCES) {
233 si.si_signo = SIGBUS;
235 si.si_code = BUS_ADRERR;
236 si.si_addr = (void __user *)address;
237 current->thread.arch.faultinfo = fi;
238 force_sig_info(SIGBUS, &si, current);
240 BUG_ON(err != -EFAULT);
241 si.si_signo = SIGSEGV;
242 si.si_addr = (void __user *) address;
243 current->thread.arch.faultinfo = fi;
244 force_sig_info(SIGSEGV, &si, current);
249 void relay_signal(int sig, struct uml_pt_regs *regs)
251 if (!UPT_IS_USER(regs)) {
253 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
254 "mount likely just ran out of space\n");
255 panic("Kernel mode signal %d", sig);
258 arch_examine_signal(sig, regs);
260 current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
261 force_sig(sig, current);
264 void bus_handler(int sig, struct uml_pt_regs *regs)
266 if (current->thread.fault_catcher != NULL)
267 UML_LONGJMP(current->thread.fault_catcher, 1);
268 else relay_signal(sig, regs);
271 void winch(int sig, struct uml_pt_regs *regs)
273 do_IRQ(WINCH_IRQ, regs);