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 down_read(&mm->mmap_sem);
101 vma = find_vma(mm, address);
104 if (vma->vm_start <= address)
106 if (!(vma->vm_flags & VM_GROWSDOWN))
108 if (expand_stack(vma, address))
111 * Ok, we have a good vm_area for this memory access, so
115 si_code = SEGV_ACCERR;
118 if (!(vma->vm_flags & VM_WRITE))
120 flags |= FAULT_FLAG_WRITE;
123 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
125 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
126 raw_smp_processor_id(),
127 current->comm, current->pid,
128 field, address, write,
129 field, regs->cp0_epc);
133 if (!(vma->vm_flags & VM_READ) &&
134 exception_epc(regs) != address) {
136 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
137 raw_smp_processor_id(),
138 current->comm, current->pid,
139 field, address, write,
140 field, regs->cp0_epc);
145 if (unlikely(!vma_is_accessible(vma)))
151 * If for any reason at all we couldn't handle the fault,
152 * make sure we exit gracefully rather than endlessly redo
155 fault = handle_mm_fault(vma, address, flags);
157 if (fault_signal_pending(fault, regs))
160 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
161 if (unlikely(fault & VM_FAULT_ERROR)) {
162 if (fault & VM_FAULT_OOM)
164 else if (fault & VM_FAULT_SIGSEGV)
166 else if (fault & VM_FAULT_SIGBUS)
170 if (flags & FAULT_FLAG_ALLOW_RETRY) {
171 if (fault & VM_FAULT_MAJOR) {
172 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
176 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
180 if (fault & VM_FAULT_RETRY) {
181 flags |= FAULT_FLAG_TRIED;
184 * No need to up_read(&mm->mmap_sem) as we would
185 * have already released it in __lock_page_or_retry
193 up_read(&mm->mmap_sem);
197 * Something tried to access memory that isn't in our memory map..
198 * Fix it, but check if it's kernel or user first..
201 up_read(&mm->mmap_sem);
203 bad_area_nosemaphore:
204 /* User mode accesses just cause a SIGSEGV */
205 if (user_mode(regs)) {
206 tsk->thread.cp0_badvaddr = address;
207 tsk->thread.error_code = write;
208 if (show_unhandled_signals &&
209 unhandled_signal(tsk, SIGSEGV) &&
210 __ratelimit(&ratelimit_state)) {
211 pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n",
213 write ? "write access to" : "read access from",
215 pr_info("epc = %0*lx in", field,
216 (unsigned long) regs->cp0_epc);
217 print_vma_addr(KERN_CONT " ", regs->cp0_epc);
219 pr_info("ra = %0*lx in", field,
220 (unsigned long) regs->regs[31]);
221 print_vma_addr(KERN_CONT " ", regs->regs[31]);
224 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
225 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
230 /* Are we prepared to handle this kernel fault? */
231 if (fixup_exception(regs)) {
232 current->thread.cp0_baduaddr = address;
237 * Oops. The kernel tried to access some bad page. We'll have to
238 * terminate things with extreme prejudice.
242 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
243 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
244 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
245 field, regs->regs[31]);
250 * We ran out of memory, call the OOM killer, and return the userspace
251 * (which will retry the fault, or kill us if we got oom-killed).
253 up_read(&mm->mmap_sem);
254 if (!user_mode(regs))
256 pagefault_out_of_memory();
260 up_read(&mm->mmap_sem);
262 /* Kernel mode? Handle exceptions or die */
263 if (!user_mode(regs))
267 * Send a sigbus, regardless of whether we were in kernel
271 printk("do_page_fault() #3: sending SIGBUS to %s for "
272 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
274 write ? "write access to" : "read access from",
276 field, (unsigned long) regs->cp0_epc,
277 field, (unsigned long) regs->regs[31]);
279 current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
280 tsk->thread.cp0_badvaddr = address;
281 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
288 * Synchronize this task's top level page-table
289 * with the 'reference' page table.
291 * Do _not_ use "tsk" here. We might be inside
292 * an interrupt in the middle of a task switch..
294 int offset = pgd_index(address);
301 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
302 pgd_k = init_mm.pgd + offset;
304 if (!pgd_present(*pgd_k))
306 set_pgd(pgd, *pgd_k);
308 p4d = p4d_offset(pgd, address);
309 p4d_k = p4d_offset(pgd_k, address);
310 if (!p4d_present(*p4d_k))
313 pud = pud_offset(p4d, address);
314 pud_k = pud_offset(p4d_k, address);
315 if (!pud_present(*pud_k))
318 pmd = pmd_offset(pud, address);
319 pmd_k = pmd_offset(pud_k, address);
320 if (!pmd_present(*pmd_k))
322 set_pmd(pmd, *pmd_k);
324 pte_k = pte_offset_kernel(pmd_k, address);
325 if (!pte_present(*pte_k))
332 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
333 unsigned long write, unsigned long address)
335 enum ctx_state prev_state;
337 prev_state = exception_enter();
338 __do_page_fault(regs, write, address);
339 exception_exit(prev_state);