2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * Copyright (C) 1995 - 2000 by Ralf Baechle
8 #include <linux/signal.h>
9 #include <linux/sched.h>
10 #include <linux/interrupt.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
18 #include <linux/smp.h>
19 #include <linux/module.h>
20 #include <linux/kprobes.h>
21 #include <linux/perf_event.h>
23 #include <asm/branch.h>
24 #include <asm/mmu_context.h>
25 #include <asm/uaccess.h>
26 #include <asm/ptrace.h>
27 #include <asm/highmem.h> /* For VMALLOC_END */
28 #include <linux/kdebug.h>
31 * This routine handles page faults. It determines the address,
32 * and the problem, and then passes it off to one of the appropriate
35 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, unsigned long write,
36 unsigned long address)
38 struct vm_area_struct * vma = NULL;
39 struct task_struct *tsk = current;
40 struct mm_struct *mm = tsk->mm;
41 const int field = sizeof(unsigned long) * 2;
44 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
47 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
48 current->comm, current->pid, field, address, write,
49 field, regs->cp0_epc);
54 * This is to notify the fault handler of the kprobes. The
55 * exception code is redundant as it is also carried in REGS,
56 * but we pass it anyhow.
58 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
59 (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
63 info.si_code = SEGV_MAPERR;
66 * We fault-in kernel-space virtual memory on-demand. The
67 * 'reference' page table is init_mm.pgd.
69 * NOTE! We MUST NOT take any locks for this case. We may
70 * be in an interrupt or a critical region, and should
71 * only copy the information from the master page table,
75 # define VMALLOC_FAULT_TARGET no_context
77 # define VMALLOC_FAULT_TARGET vmalloc_fault
80 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
81 goto VMALLOC_FAULT_TARGET;
83 if (unlikely(address >= MODULE_START && address < MODULE_END))
84 goto VMALLOC_FAULT_TARGET;
88 * If we're in an interrupt or have no user
89 * context, we must not take the fault..
91 if (in_atomic() || !mm)
92 goto bad_area_nosemaphore;
95 flags |= FAULT_FLAG_USER;
97 down_read(&mm->mmap_sem);
98 vma = find_vma(mm, address);
101 if (vma->vm_start <= address)
103 if (!(vma->vm_flags & VM_GROWSDOWN))
105 if (expand_stack(vma, address))
108 * Ok, we have a good vm_area for this memory access, so
112 info.si_code = SEGV_ACCERR;
115 if (!(vma->vm_flags & VM_WRITE))
117 flags |= FAULT_FLAG_WRITE;
120 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
122 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
123 raw_smp_processor_id(),
124 current->comm, current->pid,
125 field, address, write,
126 field, regs->cp0_epc);
130 if (!(vma->vm_flags & VM_READ)) {
132 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
133 raw_smp_processor_id(),
134 current->comm, current->pid,
135 field, address, write,
136 field, regs->cp0_epc);
141 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
147 * If for any reason at all we couldn't handle the fault,
148 * make sure we exit gracefully rather than endlessly redo
151 fault = handle_mm_fault(mm, vma, address, flags);
153 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
156 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
157 if (unlikely(fault & VM_FAULT_ERROR)) {
158 if (fault & VM_FAULT_OOM)
160 else if (fault & VM_FAULT_SIGBUS)
164 if (flags & FAULT_FLAG_ALLOW_RETRY) {
165 if (fault & VM_FAULT_MAJOR) {
166 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
170 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
174 if (fault & VM_FAULT_RETRY) {
175 flags &= ~FAULT_FLAG_ALLOW_RETRY;
176 flags |= FAULT_FLAG_TRIED;
179 * No need to up_read(&mm->mmap_sem) as we would
180 * have already released it in __lock_page_or_retry
188 up_read(&mm->mmap_sem);
192 * Something tried to access memory that isn't in our memory map..
193 * Fix it, but check if it's kernel or user first..
196 up_read(&mm->mmap_sem);
198 bad_area_nosemaphore:
199 /* User mode accesses just cause a SIGSEGV */
200 if (user_mode(regs)) {
201 tsk->thread.cp0_badvaddr = address;
202 tsk->thread.error_code = write;
204 printk("do_page_fault() #2: sending SIGSEGV to %s for "
205 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
207 write ? "write access to" : "read access from",
209 field, (unsigned long) regs->cp0_epc,
210 field, (unsigned long) regs->regs[31]);
212 info.si_signo = SIGSEGV;
214 /* info.si_code has been set above */
215 info.si_addr = (void __user *) address;
216 force_sig_info(SIGSEGV, &info, tsk);
221 /* Are we prepared to handle this kernel fault? */
222 if (fixup_exception(regs)) {
223 current->thread.cp0_baduaddr = address;
228 * Oops. The kernel tried to access some bad page. We'll have to
229 * terminate things with extreme prejudice.
233 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
234 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
235 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
236 field, regs->regs[31]);
241 * We ran out of memory, call the OOM killer, and return the userspace
242 * (which will retry the fault, or kill us if we got oom-killed).
244 up_read(&mm->mmap_sem);
245 if (!user_mode(regs))
247 pagefault_out_of_memory();
251 up_read(&mm->mmap_sem);
253 /* Kernel mode? Handle exceptions or die */
254 if (!user_mode(regs))
258 * Send a sigbus, regardless of whether we were in kernel
262 printk("do_page_fault() #3: sending SIGBUS to %s for "
263 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
265 write ? "write access to" : "read access from",
267 field, (unsigned long) regs->cp0_epc,
268 field, (unsigned long) regs->regs[31]);
270 tsk->thread.cp0_badvaddr = address;
271 info.si_signo = SIGBUS;
273 info.si_code = BUS_ADRERR;
274 info.si_addr = (void __user *) address;
275 force_sig_info(SIGBUS, &info, tsk);
282 * Synchronize this task's top level page-table
283 * with the 'reference' page table.
285 * Do _not_ use "tsk" here. We might be inside
286 * an interrupt in the middle of a task switch..
288 int offset = __pgd_offset(address);
294 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
295 pgd_k = init_mm.pgd + offset;
297 if (!pgd_present(*pgd_k))
299 set_pgd(pgd, *pgd_k);
301 pud = pud_offset(pgd, address);
302 pud_k = pud_offset(pgd_k, address);
303 if (!pud_present(*pud_k))
306 pmd = pmd_offset(pud, address);
307 pmd_k = pmd_offset(pud_k, address);
308 if (!pmd_present(*pmd_k))
310 set_pmd(pmd, *pmd_k);
312 pte_k = pte_offset_kernel(pmd_k, address);
313 if (!pte_present(*pte_k))