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/mman.h>
19 #include <linux/smp.h>
20 #include <linux/module.h>
21 #include <linux/kprobes.h>
22 #include <linux/perf_event.h>
24 #include <asm/branch.h>
25 #include <asm/mmu_context.h>
26 #include <asm/uaccess.h>
27 #include <asm/ptrace.h>
28 #include <asm/highmem.h> /* For VMALLOC_END */
29 #include <linux/kdebug.h>
32 * This routine handles page faults. It determines the address,
33 * and the problem, and then passes it off to one of the appropriate
36 static void __kprobes __do_page_fault(struct pt_regs *regs, unsigned long write,
37 unsigned long address)
39 struct vm_area_struct * vma = NULL;
40 struct task_struct *tsk = current;
41 struct mm_struct *mm = tsk->mm;
42 const int field = sizeof(unsigned long) * 2;
45 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
46 (write ? FAULT_FLAG_WRITE : 0);
49 printk("Cpu%d[%s:%d:%0*lx:%ld:%0*lx]\n", raw_smp_processor_id(),
50 current->comm, current->pid, field, address, write,
51 field, regs->cp0_epc);
56 * This is to notify the fault handler of the kprobes. The
57 * exception code is redundant as it is also carried in REGS,
58 * but we pass it anyhow.
60 if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
61 (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
65 info.si_code = SEGV_MAPERR;
68 * We fault-in kernel-space virtual memory on-demand. The
69 * 'reference' page table is init_mm.pgd.
71 * NOTE! We MUST NOT take any locks for this case. We may
72 * be in an interrupt or a critical region, and should
73 * only copy the information from the master page table,
77 # define VMALLOC_FAULT_TARGET no_context
79 # define VMALLOC_FAULT_TARGET vmalloc_fault
82 if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
83 goto VMALLOC_FAULT_TARGET;
85 if (unlikely(address >= MODULE_START && address < MODULE_END))
86 goto VMALLOC_FAULT_TARGET;
90 * If we're in an interrupt or have no user
91 * context, we must not take the fault..
93 if (in_atomic() || !mm)
94 goto bad_area_nosemaphore;
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))
119 if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
121 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
122 raw_smp_processor_id(),
123 current->comm, current->pid,
124 field, address, write,
125 field, regs->cp0_epc);
129 if (!(vma->vm_flags & VM_READ)) {
131 pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] RI violation\n",
132 raw_smp_processor_id(),
133 current->comm, current->pid,
134 field, address, write,
135 field, regs->cp0_epc);
140 if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
146 * If for any reason at all we couldn't handle the fault,
147 * make sure we exit gracefully rather than endlessly redo
150 fault = handle_mm_fault(mm, vma, address, flags);
152 if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
155 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
156 if (unlikely(fault & VM_FAULT_ERROR)) {
157 if (fault & VM_FAULT_OOM)
159 else if (fault & VM_FAULT_SIGBUS)
163 if (flags & FAULT_FLAG_ALLOW_RETRY) {
164 if (fault & VM_FAULT_MAJOR) {
165 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
169 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
173 if (fault & VM_FAULT_RETRY) {
174 flags &= ~FAULT_FLAG_ALLOW_RETRY;
175 flags |= FAULT_FLAG_TRIED;
178 * No need to up_read(&mm->mmap_sem) as we would
179 * have already released it in __lock_page_or_retry
187 up_read(&mm->mmap_sem);
191 * Something tried to access memory that isn't in our memory map..
192 * Fix it, but check if it's kernel or user first..
195 up_read(&mm->mmap_sem);
197 bad_area_nosemaphore:
198 /* User mode accesses just cause a SIGSEGV */
199 if (user_mode(regs)) {
200 tsk->thread.cp0_badvaddr = address;
201 tsk->thread.error_code = write;
203 printk("do_page_fault() #2: sending SIGSEGV to %s for "
204 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
206 write ? "write access to" : "read access from",
208 field, (unsigned long) regs->cp0_epc,
209 field, (unsigned long) regs->regs[31]);
211 info.si_signo = SIGSEGV;
213 /* info.si_code has been set above */
214 info.si_addr = (void __user *) address;
215 force_sig_info(SIGSEGV, &info, tsk);
220 /* Are we prepared to handle this kernel fault? */
221 if (fixup_exception(regs)) {
222 current->thread.cp0_baduaddr = address;
227 * Oops. The kernel tried to access some bad page. We'll have to
228 * terminate things with extreme prejudice.
232 printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
233 "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
234 raw_smp_processor_id(), field, address, field, regs->cp0_epc,
235 field, regs->regs[31]);
240 * We ran out of memory, call the OOM killer, and return the userspace
241 * (which will retry the fault, or kill us if we got oom-killed).
243 up_read(&mm->mmap_sem);
244 if (!user_mode(regs))
246 pagefault_out_of_memory();
250 up_read(&mm->mmap_sem);
252 /* Kernel mode? Handle exceptions or die */
253 if (!user_mode(regs))
257 * Send a sigbus, regardless of whether we were in kernel
261 printk("do_page_fault() #3: sending SIGBUS to %s for "
262 "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
264 write ? "write access to" : "read access from",
266 field, (unsigned long) regs->cp0_epc,
267 field, (unsigned long) regs->regs[31]);
269 tsk->thread.cp0_badvaddr = address;
270 info.si_signo = SIGBUS;
272 info.si_code = BUS_ADRERR;
273 info.si_addr = (void __user *) address;
274 force_sig_info(SIGBUS, &info, tsk);
281 * Synchronize this task's top level page-table
282 * with the 'reference' page table.
284 * Do _not_ use "tsk" here. We might be inside
285 * an interrupt in the middle of a task switch..
287 int offset = __pgd_offset(address);
293 pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
294 pgd_k = init_mm.pgd + offset;
296 if (!pgd_present(*pgd_k))
298 set_pgd(pgd, *pgd_k);
300 pud = pud_offset(pgd, address);
301 pud_k = pud_offset(pgd_k, address);
302 if (!pud_present(*pud_k))
305 pmd = pmd_offset(pud, address);
306 pmd_k = pmd_offset(pud_k, address);
307 if (!pmd_present(*pmd_k))
309 set_pmd(pmd, *pmd_k);
311 pte_k = pte_offset_kernel(pmd_k, address);
312 if (!pte_present(*pte_k))
319 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
320 unsigned long write, unsigned long address)
322 enum ctx_state prev_state;
324 prev_state = exception_enter();
325 __do_page_fault(regs, write, address);
326 exception_exit(prev_state);