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 <linux/uaccess.h>
11 #include <asm/current.h>
12 #include <asm/pgtable.h>
13 #include <asm/tlbflush.h>
15 #include <as-layout.h>
16 #include <kern_util.h>
21 * Note this is constrained to return 0, -EFAULT, -EACCESS, -ENOMEM by
24 int handle_page_fault(unsigned long address, unsigned long ip,
25 int is_write, int is_user, int *code_out)
27 struct mm_struct *mm = current->mm;
28 struct vm_area_struct *vma;
34 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
36 *code_out = SEGV_MAPERR;
39 * If the fault was with pagefaults disabled, don't take the fault, just
42 if (faulthandler_disabled())
46 flags |= FAULT_FLAG_USER;
48 down_read(&mm->mmap_sem);
49 vma = find_vma(mm, address);
52 else if (vma->vm_start <= address)
54 else if (!(vma->vm_flags & VM_GROWSDOWN))
56 else if (is_user && !ARCH_IS_STACKGROW(address))
58 else if (expand_stack(vma, address))
62 *code_out = SEGV_ACCERR;
64 if (!(vma->vm_flags & VM_WRITE))
66 flags |= FAULT_FLAG_WRITE;
68 /* Don't require VM_READ|VM_EXEC for write faults! */
69 if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
76 fault = handle_mm_fault(mm, vma, address, flags);
78 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
81 if (unlikely(fault & VM_FAULT_ERROR)) {
82 if (fault & VM_FAULT_OOM) {
84 } else if (fault & VM_FAULT_SIGSEGV) {
86 } else if (fault & VM_FAULT_SIGBUS) {
92 if (flags & FAULT_FLAG_ALLOW_RETRY) {
93 if (fault & VM_FAULT_MAJOR)
97 if (fault & VM_FAULT_RETRY) {
98 flags &= ~FAULT_FLAG_ALLOW_RETRY;
99 flags |= FAULT_FLAG_TRIED;
105 pgd = pgd_offset(mm, address);
106 pud = pud_offset(pgd, address);
107 pmd = pmd_offset(pud, address);
108 pte = pte_offset_kernel(pmd, address);
109 } while (!pte_present(*pte));
112 * The below warning was added in place of
113 * pte_mkyoung(); if (is_write) pte_mkdirty();
114 * If it's triggered, we'd see normally a hang here (a clean pte is
115 * marked read-only to emulate the dirty bit).
116 * However, the generic code can mark a PTE writable but clean on a
117 * concurrent read fault, triggering this harmlessly. So comment it out.
120 WARN_ON(!pte_young(*pte) || (is_write && !pte_dirty(*pte)));
122 flush_tlb_page(vma, address);
124 up_read(&mm->mmap_sem);
130 * We ran out of memory, call the OOM killer, and return the userspace
131 * (which will retry the fault, or kill us if we got oom-killed).
133 up_read(&mm->mmap_sem);
135 goto out_nosemaphore;
136 pagefault_out_of_memory();
139 EXPORT_SYMBOL(handle_page_fault);
141 static void show_segv_info(struct uml_pt_regs *regs)
143 struct task_struct *tsk = current;
144 struct faultinfo *fi = UPT_FAULTINFO(regs);
146 if (!unhandled_signal(tsk, SIGSEGV))
149 if (!printk_ratelimit())
152 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %x",
153 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
154 tsk->comm, task_pid_nr(tsk), FAULT_ADDRESS(*fi),
155 (void *)UPT_IP(regs), (void *)UPT_SP(regs),
158 print_vma_addr(KERN_CONT " in ", UPT_IP(regs));
159 printk(KERN_CONT "\n");
162 static void bad_segv(struct faultinfo fi, unsigned long ip)
166 si.si_signo = SIGSEGV;
167 si.si_code = SEGV_ACCERR;
168 si.si_addr = (void __user *) FAULT_ADDRESS(fi);
169 current->thread.arch.faultinfo = fi;
170 force_sig_info(SIGSEGV, &si, current);
173 void fatal_sigsegv(void)
175 force_sigsegv(SIGSEGV, current);
176 do_signal(¤t->thread.regs);
178 * This is to tell gcc that we're not returning - do_signal
179 * can, in general, return, but in this case, it's not, since
180 * we just got a fatal SIGSEGV queued.
185 void segv_handler(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
187 struct faultinfo * fi = UPT_FAULTINFO(regs);
189 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
190 show_segv_info(regs);
191 bad_segv(*fi, UPT_IP(regs));
194 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
198 * We give a *copy* of the faultinfo in the regs to segv.
199 * This must be done, since nesting SEGVs could overwrite
200 * the info in the regs. A pointer to the info then would
203 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
204 struct uml_pt_regs *regs)
209 int is_write = FAULT_WRITE(fi);
210 unsigned long address = FAULT_ADDRESS(fi);
212 if (!is_user && regs)
213 current->thread.segv_regs = container_of(regs, struct pt_regs, regs);
215 if (!is_user && (address >= start_vm) && (address < end_vm)) {
216 flush_tlb_kernel_vm();
219 else if (current->mm == NULL) {
220 show_regs(container_of(regs, struct pt_regs, regs));
221 panic("Segfault with no mm");
223 else if (!is_user && address > PAGE_SIZE && address < TASK_SIZE) {
224 show_regs(container_of(regs, struct pt_regs, regs));
225 panic("Kernel tried to access user memory at addr 0x%lx, ip 0x%lx",
229 if (SEGV_IS_FIXABLE(&fi))
230 err = handle_page_fault(address, ip, is_write, is_user,
235 * A thread accessed NULL, we get a fault, but CR2 is invalid.
236 * This code is used in __do_copy_from_user() of TT mode.
237 * XXX tt mode is gone, so maybe this isn't needed any more
242 catcher = current->thread.fault_catcher;
245 else if (catcher != NULL) {
246 current->thread.fault_addr = (void *) address;
247 UML_LONGJMP(catcher, 1);
249 else if (current->thread.fault_addr != NULL)
250 panic("fault_addr set but no fault catcher");
251 else if (!is_user && arch_fixup(ip, regs))
255 show_regs(container_of(regs, struct pt_regs, regs));
256 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
260 show_segv_info(regs);
262 if (err == -EACCES) {
263 si.si_signo = SIGBUS;
265 si.si_code = BUS_ADRERR;
266 si.si_addr = (void __user *)address;
267 current->thread.arch.faultinfo = fi;
268 force_sig_info(SIGBUS, &si, current);
270 BUG_ON(err != -EFAULT);
271 si.si_signo = SIGSEGV;
272 si.si_addr = (void __user *) address;
273 current->thread.arch.faultinfo = fi;
274 force_sig_info(SIGSEGV, &si, current);
279 current->thread.segv_regs = NULL;
284 void relay_signal(int sig, struct siginfo *si, struct uml_pt_regs *regs)
286 struct faultinfo *fi;
287 struct siginfo clean_si;
289 if (!UPT_IS_USER(regs)) {
291 printk(KERN_ERR "Bus error - the host /dev/shm or /tmp "
292 "mount likely just ran out of space\n");
293 panic("Kernel mode signal %d", sig);
296 arch_examine_signal(sig, regs);
298 memset(&clean_si, 0, sizeof(clean_si));
299 clean_si.si_signo = si->si_signo;
300 clean_si.si_errno = si->si_errno;
301 clean_si.si_code = si->si_code;
308 fi = UPT_FAULTINFO(regs);
309 clean_si.si_addr = (void __user *) FAULT_ADDRESS(*fi);
310 current->thread.arch.faultinfo = *fi;
311 #ifdef __ARCH_SI_TRAPNO
312 clean_si.si_trapno = si->si_trapno;
316 printk(KERN_ERR "Attempted to relay unknown signal %d (si_code = %d)\n",
320 force_sig_info(sig, &clean_si, current);
323 void bus_handler(int sig, struct siginfo *si, struct uml_pt_regs *regs)
325 if (current->thread.fault_catcher != NULL)
326 UML_LONGJMP(current->thread.fault_catcher, 1);
328 relay_signal(sig, si, regs);
331 void winch(int sig, struct siginfo *unused_si, struct uml_pt_regs *regs)
333 do_IRQ(WINCH_IRQ, regs);