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/module.h>
22 #include <linux/kprobes.h>
23 #include <linux/perf_event.h>
24 #include <linux/uaccess.h>
26 #include <asm/branch.h>
27 #include <asm/mmu_context.h>
28 #include <asm/ptrace.h>
29 #include <asm/highmem.h> /* For VMALLOC_END */
30 #include <linux/kdebug.h>
32 int show_unhandled_signals = 1;
35 * This routine handles page faults. It determines the address,
36 * and the problem, and then passes it off to one of the appropriate
39 static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write,
40 unsigned long address)
42 struct vm_area_struct * vma = NULL;
43 struct task_struct *tsk = current;
44 struct mm_struct *mm = tsk->mm;
45 const int field = sizeof(unsigned long) * 2;
48 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
50 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
53 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
54 current->comm, current->pid, field, address, write,
55 field, regs->cp0_epc);
60 * This is to notify the fault handler of the kprobes. The
61 * exception code is redundant as it is also carried in REGS,
62 * but we pass it anyhow.
64 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
65 (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
69 info.si_code = SEGV_MAPERR;
72 * We fault-in kernel-space virtual memory on-demand. The
73 * 'reference' page table is init_mm.pgd.
75 * NOTE! We MUST NOT take any locks for this case. We may
76 * be in an interrupt or a critical region, and should
77 * only copy the information from the master page table,
81 # define VMALLOC_FAULT_TARGET no_context
83 # define VMALLOC_FAULT_TARGET vmalloc_fault
86 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
87 goto VMALLOC_FAULT_TARGET;
89 if (unlikely(address >= MODULE_START && address < MODULE_END))
90 goto VMALLOC_FAULT_TARGET;
94 * If we're in an interrupt or have no user
95 * context, we must not take the fault..
97 if (faulthandler_disabled() || !mm)
98 goto bad_area_nosemaphore;
101 flags |= FAULT_FLAG_USER;
103 down_read(&mm->mmap_sem);
104 vma = find_vma(mm, address);
107 if (vma->vm_start <= address)
109 if (!(vma->vm_flags & VM_GROWSDOWN))
111 if (expand_stack(vma, address))
114 * Ok, we have a good vm_area for this memory access, so
118 info.si_code = SEGV_ACCERR;
121 if (!(vma->vm_flags & VM_WRITE))
123 flags |= FAULT_FLAG_WRITE;
126 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
128 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
129 raw_smp_processor_id(),
130 current->comm, current->pid,
131 field, address, write,
132 field, regs->cp0_epc);
136 if (!(vma->vm_flags & VM_READ) &&
137 exception_epc(regs) != address) {
139 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
140 raw_smp_processor_id(),
141 current->comm, current->pid,
142 field, address, write,
143 field, regs->cp0_epc);
148 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
154 * If for any reason at all we couldn't handle the fault,
155 * make sure we exit gracefully rather than endlessly redo
158 fault = handle_mm_fault(mm, vma, address, flags);
160 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
163 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
164 if (unlikely(fault & VM_FAULT_ERROR)) {
165 if (fault & VM_FAULT_OOM)
167 else if (fault & VM_FAULT_SIGSEGV)
169 else if (fault & VM_FAULT_SIGBUS)
173 if (flags & FAULT_FLAG_ALLOW_RETRY) {
174 if (fault & VM_FAULT_MAJOR) {
175 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
179 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
183 if (fault & VM_FAULT_RETRY) {
184 flags &= ~FAULT_FLAG_ALLOW_RETRY;
185 flags |= FAULT_FLAG_TRIED;
188 * No need to up_read(&mm->mmap_sem) as we would
189 * have already released it in __lock_page_or_retry
197 up_read(&mm->mmap_sem);
201 * Something tried to access memory that isn't in our memory map..
202 * Fix it, but check if it's kernel or user first..
205 up_read(&mm->mmap_sem);
207 bad_area_nosemaphore:
208 /* User mode accesses just cause a SIGSEGV */
209 if (user_mode(regs)) {
210 tsk->thread.cp0_badvaddr = address;
211 tsk->thread.error_code = write;
212 if (show_unhandled_signals &&
213 unhandled_signal(tsk, SIGSEGV) &&
214 __ratelimit(&ratelimit_state)) {
215 pr_info("\ndo_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx",
217 write ? "write access to" : "read access from",
219 pr_info("epc = %0*lx in", field,
220 (unsigned long) regs->cp0_epc);
221 print_vma_addr(" ", regs->cp0_epc);
222 pr_info("ra = %0*lx in", field,
223 (unsigned long) regs->regs[31]);
224 print_vma_addr(" ", regs->regs[31]);
227 info.si_signo = SIGSEGV;
229 /* info.si_code has been set above */
230 info.si_addr = (void __user *) address;
231 force_sig_info(SIGSEGV, &info, tsk);
236 /* Are we prepared to handle this kernel fault? */
237 if (fixup_exception(regs)) {
238 current->thread.cp0_baduaddr = address;
243 * Oops. The kernel tried to access some bad page. We'll have to
244 * terminate things with extreme prejudice.
248 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
249 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
250 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
251 field, regs->regs[31]);
256 * We ran out of memory, call the OOM killer, and return the userspace
257 * (which will retry the fault, or kill us if we got oom-killed).
259 up_read(&mm->mmap_sem);
260 if (!user_mode(regs))
262 pagefault_out_of_memory();
266 up_read(&mm->mmap_sem);
268 /* Kernel mode? Handle exceptions or die */
269 if (!user_mode(regs))
273 * Send a sigbus, regardless of whether we were in kernel
277 printk("do_page_fault() #3: sending SIGBUS to %s for "
278 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
280 write ? "write access to" : "read access from",
282 field, (unsigned long) regs->cp0_epc,
283 field, (unsigned long) regs->regs[31]);
285 tsk->thread.cp0_badvaddr = address;
286 info.si_signo = SIGBUS;
288 info.si_code = BUS_ADRERR;
289 info.si_addr = (void __user *) address;
290 force_sig_info(SIGBUS, &info, tsk);
297 * Synchronize this task's top level page-table
298 * with the 'reference' page table.
300 * Do _not_ use "tsk" here. We might be inside
301 * an interrupt in the middle of a task switch..
303 int offset = __pgd_offset(address);
309 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
310 pgd_k = init_mm.pgd + offset;
312 if (!pgd_present(*pgd_k))
314 set_pgd(pgd, *pgd_k);
316 pud = pud_offset(pgd, address);
317 pud_k = pud_offset(pgd_k, address);
318 if (!pud_present(*pud_k))
321 pmd = pmd_offset(pud, address);
322 pmd_k = pmd_offset(pud_k, address);
323 if (!pmd_present(*pmd_k))
325 set_pmd(pmd, *pmd_k);
327 pte_k = pte_offset_kernel(pmd_k, address);
328 if (!pte_present(*pte_k))
335 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
336 unsigned long write, unsigned long address)
338 enum ctx_state prev_state;
340 prev_state = exception_enter();
341 __do_page_fault(regs, write, address);
342 exception_exit(prev_state);