1/*
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
4 * for more details.
5 *
6 * Copyright (C) 1995 - 2000 by Ralf Baechle
7 */
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>
19#include <linux/mm.h>
20#include <linux/smp.h>
21#include <linux/kprobes.h>
22#include <linux/perf_event.h>
23#include <linux/uaccess.h>
24
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 <asm/traps.h>
30#include <linux/kdebug.h>
31
32int show_unhandled_signals = 1;
33
34/*
35 * This routine handles page faults.  It determines the address,
36 * and the problem, and then passes it off to one of the appropriate
37 * routines.
38 */
39static void __do_page_fault(struct pt_regs *regs, unsigned long write,
40	unsigned long address)
41{
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;
46	int si_code;
47	vm_fault_t fault;
48	unsigned int flags = FAULT_FLAG_DEFAULT;
49
50	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
51
52#if 0
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);
56#endif
57
58#ifdef CONFIG_KPROBES
59	/*
60	 * This is to notify the fault handler of the kprobes.
61	 */
62	if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
63		       current->thread.trap_nr, SIGSEGV) == NOTIFY_STOP)
64		return;
65#endif
66
67	si_code = SEGV_MAPERR;
68
69	/*
70	 * We fault-in kernel-space virtual memory on-demand. The
71	 * 'reference' page table is init_mm.pgd.
72	 *
73	 * NOTE! We MUST NOT take any locks for this case. We may
74	 * be in an interrupt or a critical region, and should
75	 * only copy the information from the master page table,
76	 * nothing more.
77	 */
78#ifdef CONFIG_64BIT
79# define VMALLOC_FAULT_TARGET no_context
80#else
81# define VMALLOC_FAULT_TARGET vmalloc_fault
82#endif
83
84	if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
85		goto VMALLOC_FAULT_TARGET;
86#ifdef MODULE_START
87	if (unlikely(address >= MODULE_START && address < MODULE_END))
88		goto VMALLOC_FAULT_TARGET;
89#endif
90
91	/*
92	 * If we're in an interrupt or have no user
93	 * context, we must not take the fault..
94	 */
95	if (faulthandler_disabled() || !mm)
96		goto bad_area_nosemaphore;
97
98	if (user_mode(regs))
99		flags |= FAULT_FLAG_USER;
100
101	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
102retry:
103	vma = lock_mm_and_find_vma(mm, address, regs);
104	if (!vma)
105		goto bad_area_nosemaphore;
106/*
107 * Ok, we have a good vm_area for this memory access, so
108 * we can handle it..
109 */
110	si_code = SEGV_ACCERR;
111
112	if (write) {
113		if (!(vma->vm_flags & VM_WRITE))
114			goto bad_area;
115		flags |= FAULT_FLAG_WRITE;
116	} else {
117		if (cpu_has_rixi) {
118			if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
119#if 0
120				pr_notice("Cpu%d[%s:%d:%0*lx:%ld:%0*lx] XI violation\n",
121					  raw_smp_processor_id(),
122					  current->comm, current->pid,
123					  field, address, write,
124					  field, regs->cp0_epc);
125#endif
126				goto bad_area;
127			}
128			if (!(vma->vm_flags & VM_READ) &&
129			    exception_epc(regs) != address) {
130#if 0
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);
136#endif
137				goto bad_area;
138			}
139		} else {
140			if (unlikely(!vma_is_accessible(vma)))
141				goto bad_area;
142		}
143	}
144
145	/*
146	 * If for any reason at all we couldn't handle the fault,
147	 * make sure we exit gracefully rather than endlessly redo
148	 * the fault.
149	 */
150	fault = handle_mm_fault(vma, address, flags, regs);
151
152	if (fault_signal_pending(fault, regs)) {
153		if (!user_mode(regs))
154			goto no_context;
155		return;
156	}
157
158	/* The fault is fully completed (including releasing mmap lock) */
159	if (fault & VM_FAULT_COMPLETED)
160		return;
161
162	if (unlikely(fault & VM_FAULT_ERROR)) {
163		if (fault & VM_FAULT_OOM)
164			goto out_of_memory;
165		else if (fault & VM_FAULT_SIGSEGV)
166			goto bad_area;
167		else if (fault & VM_FAULT_SIGBUS)
168			goto do_sigbus;
169		BUG();
170	}
171
172	if (fault & VM_FAULT_RETRY) {
173		flags |= FAULT_FLAG_TRIED;
174
175		/*
176		 * No need to mmap_read_unlock(mm) as we would
177		 * have already released it in __lock_page_or_retry
178		 * in mm/filemap.c.
179		 */
180
181		goto retry;
182	}
183
184	mmap_read_unlock(mm);
185	return;
186
187/*
188 * Something tried to access memory that isn't in our memory map..
189 * Fix it, but check if it's kernel or user first..
190 */
191bad_area:
192	mmap_read_unlock(mm);
193
194bad_area_nosemaphore:
195	/* User mode accesses just cause a SIGSEGV */
196	if (user_mode(regs)) {
197		tsk->thread.cp0_badvaddr = address;
198		tsk->thread.error_code = write;
199		if (show_unhandled_signals &&
200		    unhandled_signal(tsk, SIGSEGV) &&
201		    __ratelimit(&ratelimit_state)) {
202			pr_info("do_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx\n",
203				tsk->comm,
204				write ? "write access to" : "read access from",
205				field, address);
206			pr_info("epc = %0*lx in", field,
207				(unsigned long) regs->cp0_epc);
208			print_vma_addr(KERN_CONT " ", regs->cp0_epc);
209			pr_cont("\n");
210			pr_info("ra  = %0*lx in", field,
211				(unsigned long) regs->regs[31]);
212			print_vma_addr(KERN_CONT " ", regs->regs[31]);
213			pr_cont("\n");
214		}
215		current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
216		force_sig_fault(SIGSEGV, si_code, (void __user *)address);
217		return;
218	}
219
220no_context:
221	/* Are we prepared to handle this kernel fault?	 */
222	if (fixup_exception(regs)) {
223		current->thread.cp0_baduaddr = address;
224		return;
225	}
226
227	/*
228	 * Oops. The kernel tried to access some bad page. We'll have to
229	 * terminate things with extreme prejudice.
230	 */
231	bust_spinlocks(1);
232
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]);
237	die("Oops", regs);
238
239out_of_memory:
240	/*
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).
243	 */
244	mmap_read_unlock(mm);
245	if (!user_mode(regs))
246		goto no_context;
247	pagefault_out_of_memory();
248	return;
249
250do_sigbus:
251	mmap_read_unlock(mm);
252
253	/* Kernel mode? Handle exceptions or die */
254	if (!user_mode(regs))
255		goto no_context;
256
257	/*
258	 * Send a sigbus, regardless of whether we were in kernel
259	 * or user mode.
260	 */
261#if 0
262	printk("do_page_fault() #3: sending SIGBUS to %s for "
263	       "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
264	       tsk->comm,
265	       write ? "write access to" : "read access from",
266	       field, address,
267	       field, (unsigned long) regs->cp0_epc,
268	       field, (unsigned long) regs->regs[31]);
269#endif
270	current->thread.trap_nr = (regs->cp0_cause >> 2) & 0x1f;
271	tsk->thread.cp0_badvaddr = address;
272	force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
273
274	return;
275#ifndef CONFIG_64BIT
276vmalloc_fault:
277	{
278		/*
279		 * Synchronize this task's top level page-table
280		 * with the 'reference' page table.
281		 *
282		 * Do _not_ use "tsk" here. We might be inside
283		 * an interrupt in the middle of a task switch..
284		 */
285		int offset = pgd_index(address);
286		pgd_t *pgd, *pgd_k;
287		p4d_t *p4d, *p4d_k;
288		pud_t *pud, *pud_k;
289		pmd_t *pmd, *pmd_k;
290		pte_t *pte_k;
291
292		pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
293		pgd_k = init_mm.pgd + offset;
294
295		if (!pgd_present(*pgd_k))
296			goto no_context;
297		set_pgd(pgd, *pgd_k);
298
299		p4d = p4d_offset(pgd, address);
300		p4d_k = p4d_offset(pgd_k, address);
301		if (!p4d_present(*p4d_k))
302			goto no_context;
303
304		pud = pud_offset(p4d, address);
305		pud_k = pud_offset(p4d_k, address);
306		if (!pud_present(*pud_k))
307			goto no_context;
308
309		pmd = pmd_offset(pud, address);
310		pmd_k = pmd_offset(pud_k, address);
311		if (!pmd_present(*pmd_k))
312			goto no_context;
313		set_pmd(pmd, *pmd_k);
314
315		pte_k = pte_offset_kernel(pmd_k, address);
316		if (!pte_present(*pte_k))
317			goto no_context;
318		return;
319	}
320#endif
321}
322NOKPROBE_SYMBOL(__do_page_fault);
323
324asmlinkage void do_page_fault(struct pt_regs *regs,
325	unsigned long write, unsigned long address)
326{
327	enum ctx_state prev_state;
328
329	prev_state = exception_enter();
330	__do_page_fault(regs, write, address);
331	exception_exit(prev_state);
332}
333NOKPROBE_SYMBOL(do_page_fault);
334