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/context_tracking.h>
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/interrupt.h>
12 #include <linux/kernel.h>
13 #include <linux/errno.h>
14 #include <linux/string.h>
15 #include <linux/types.h>
16 #include <linux/ptrace.h>
17 #include <linux/ratelimit.h>
18 #include <linux/mman.h>
20 #include <linux/smp.h>
21 #include <linux/kprobes.h>
22 #include <linux/perf_event.h>
23 #include <linux/uaccess.h>
25 #include <asm/branch.h>
26 #include <asm/mmu_context.h>
27 #include <asm/ptrace.h>
28 #include <asm/highmem.h> /* For VMALLOC_END */
29 #include <linux/kdebug.h>
31 int show_unhandled_signals = 1;
34 * This routine handles page faults. It determines the address,
35 * and the problem, and then passes it off to one of the appropriate
38 static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write,
39 unsigned long address)
41 struct vm_area_struct * vma = NULL;
42 struct task_struct *tsk = current;
43 struct mm_struct *mm = tsk->mm;
44 const int field = sizeof(unsigned long) * 2;
47 unsigned int flags = FAULT_FLAG_DEFAULT;
49 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
52 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
53 current->comm, current->pid, field, address, write,
54 field, regs->cp0_epc);
59 * This is to notify the fault handler of the kprobes.
61 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
62 current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
66 si_code = SEGV_MAPERR;
69 * We fault-in kernel-space virtual memory on-demand. The
70 * 'reference' page table is init_mm.pgd.
72 * NOTE! We MUST NOT take any locks for this case. We may
73 * be in an interrupt or a critical region, and should
74 * only copy the information from the master page table,
78 # define VMALLOC_FAULT_TARGET no_context
80 # define VMALLOC_FAULT_TARGET vmalloc_fault
83 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
84 goto VMALLOC_FAULT_TARGET;
86 if (unlikely(address >= MODULE_START && address < MODULE_END))
87 goto VMALLOC_FAULT_TARGET;
91 * If we're in an interrupt or have no user
92 * context, we must not take the fault..
94 if (faulthandler_disabled() || !mm)
95 goto bad_area_nosemaphore;
98 flags |= FAULT_FLAG_USER;
100 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
103 vma = find_vma(mm, address);
106 if (vma->vm_start <= address)
108 if (!(vma->vm_flags & VM_GROWSDOWN))
110 if (expand_stack(vma, address))
113 * Ok, we have a good vm_area for this memory access, so
117 si_code = SEGV_ACCERR;
120 if (!(vma->vm_flags & VM_WRITE))
122 flags |= FAULT_FLAG_WRITE;
125 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
127 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
128 raw_smp_processor_id(),
129 current->comm, current->pid,
130 field, address, write,
131 field, regs->cp0_epc);
135 if (!(vma->vm_flags & VM_READ) &&
136 exception_epc(regs) != address) {
138 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
139 raw_smp_processor_id(),
140 current->comm, current->pid,
141 field, address, write,
142 field, regs->cp0_epc);
147 if (unlikely(!vma_is_accessible(vma)))
153 * If for any reason at all we couldn't handle the fault,
154 * make sure we exit gracefully rather than endlessly redo
157 fault = handle_mm_fault(vma, address, flags, regs);
159 if (fault_signal_pending(fault, regs)) {
160 if (!user_mode(regs))
165 if (unlikely(fault & VM_FAULT_ERROR)) {
166 if (fault & VM_FAULT_OOM)
168 else if (fault & VM_FAULT_SIGSEGV)
170 else if (fault & VM_FAULT_SIGBUS)
174 if (flags & FAULT_FLAG_ALLOW_RETRY) {
175 if (fault & VM_FAULT_RETRY) {
176 flags |= FAULT_FLAG_TRIED;
179 * No need to mmap_read_unlock(mm) as we would
180 * have already released it in __lock_page_or_retry
188 mmap_read_unlock(mm);
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 mmap_read_unlock(mm);
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;
203 if (show_unhandled_signals &&
204 unhandled_signal(tsk, SIGSEGV) &&
205 __ratelimit(&ratelimit_state)) {
206 pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n",
208 write ? "write access to" : "read access from",
210 pr_info("epc = %0*lx in", field,
211 (unsigned long) regs->cp0_epc);
212 print_vma_addr(KERN_CONT " ", regs->cp0_epc);
214 pr_info("ra = %0*lx in", field,
215 (unsigned long) regs->regs[31]);
216 print_vma_addr(KERN_CONT " ", regs->regs[31]);
219 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
220 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
225 /* Are we prepared to handle this kernel fault? */
226 if (fixup_exception(regs)) {
227 current->thread.cp0_baduaddr = address;
232 * Oops. The kernel tried to access some bad page. We'll have to
233 * terminate things with extreme prejudice.
237 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
238 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
239 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
240 field, regs->regs[31]);
245 * We ran out of memory, call the OOM killer, and return the userspace
246 * (which will retry the fault, or kill us if we got oom-killed).
248 mmap_read_unlock(mm);
249 if (!user_mode(regs))
251 pagefault_out_of_memory();
255 mmap_read_unlock(mm);
257 /* Kernel mode? Handle exceptions or die */
258 if (!user_mode(regs))
262 * Send a sigbus, regardless of whether we were in kernel
266 printk("do_page_fault() #3: sending SIGBUS to %s for "
267 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
269 write ? "write access to" : "read access from",
271 field, (unsigned long) regs->cp0_epc,
272 field, (unsigned long) regs->regs[31]);
274 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
275 tsk->thread.cp0_badvaddr = address;
276 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
283 * Synchronize this task's top level page-table
284 * with the 'reference' page table.
286 * Do _not_ use "tsk" here. We might be inside
287 * an interrupt in the middle of a task switch..
289 int offset = pgd_index(address);
296 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
297 pgd_k = init_mm.pgd + offset;
299 if (!pgd_present(*pgd_k))
301 set_pgd(pgd, *pgd_k);
303 p4d = p4d_offset(pgd, address);
304 p4d_k = p4d_offset(pgd_k, address);
305 if (!p4d_present(*p4d_k))
308 pud = pud_offset(p4d, address);
309 pud_k = pud_offset(p4d_k, address);
310 if (!pud_present(*pud_k))
313 pmd = pmd_offset(pud, address);
314 pmd_k = pmd_offset(pud_k, address);
315 if (!pmd_present(*pmd_k))
317 set_pmd(pmd, *pmd_k);
319 pte_k = pte_offset_kernel(pmd_k, address);
320 if (!pte_present(*pte_k))
327 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
328 unsigned long write, unsigned long address)
330 enum ctx_state prev_state;
332 prev_state = exception_enter();
333 __do_page_fault(regs, write, address);
334 exception_exit(prev_state);