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 ? FAULT_FLAG_WRITE : 0);
69 if (unlikely(fault & VM_FAULT_ERROR)) {
70 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);
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).
111 up_read(&mm->mmap_sem);
112 pagefault_out_of_memory();
116 static void bad_segv(struct faultinfo fi, unsigned long ip)
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);
127 void fatal_sigsegv(void)
129 force_sigsegv(SIGSEGV, current);
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.
139 void segv_handler(int sig, struct uml_pt_regs *regs)
141 struct faultinfo * fi = UPT_FAULTINFO(regs);
143 if (UPT_IS_USER(regs) && !SEGV_IS_FIXABLE(fi)) {
144 bad_segv(*fi, UPT_IP(regs));
147 segv(*fi, UPT_IP(regs), UPT_IS_USER(regs), regs);
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
156 unsigned long segv(struct faultinfo fi, unsigned long ip, int is_user,
157 struct uml_pt_regs *regs)
162 int is_write = FAULT_WRITE(fi);
163 unsigned long address = FAULT_ADDRESS(fi);
165 if (!is_user && (address >= start_vm) && (address < end_vm)) {
166 flush_tlb_kernel_vm();
169 else if (current->mm == NULL) {
170 show_regs(container_of(regs, struct pt_regs, regs));
171 panic("Segfault with no mm");
174 if (SEGV_IS_FIXABLE(&fi) || SEGV_MAYBE_FIXABLE(&fi))
175 err = handle_page_fault(address, ip, is_write, is_user,
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
187 catcher = current->thread.fault_catcher;
190 else if (catcher != NULL) {
191 current->thread.fault_addr = (void *) address;
192 UML_LONGJMP(catcher, 1);
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))
200 show_regs(container_of(regs, struct pt_regs, regs));
201 panic("Kernel mode fault at addr 0x%lx, ip 0x%lx",
205 if (err == -EACCES) {
206 si.si_signo = SIGBUS;
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);
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);
222 void relay_signal(int sig, struct uml_pt_regs *regs)
224 if (!UPT_IS_USER(regs)) {
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);
231 arch_examine_signal(sig, regs);
233 current->thread.arch.faultinfo = *UPT_FAULTINFO(regs);
234 force_sig(sig, current);
237 void bus_handler(int sig, struct uml_pt_regs *regs)
239 if (current->thread.fault_catcher != NULL)
240 UML_LONGJMP(current->thread.fault_catcher, 1);
241 else relay_signal(sig, regs);
244 void winch(int sig, struct uml_pt_regs *regs)
246 do_IRQ(WINCH_IRQ, regs);