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/module.h>
22#include <linux/kprobes.h>
23#include <linux/perf_event.h>
24
25#include <asm/branch.h>
26#include <asm/mmu_context.h>
27#include <asm/uaccess.h>
28#include <asm/ptrace.h>
29#include <asm/highmem.h>		/* For VMALLOC_END */
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 __kprobes __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	siginfo_t info;
47	int fault;
48	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
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.	The
61	 * exception code is redundant as it is also carried in REGS,
62	 * but we pass it anyhow.
63	 */
64	if (notify_die(DIE_PAGE_FAULT, "page fault", regs, -1,
65		       (regs->cp0_cause >> 2) & 0x1f, SIGSEGV) == NOTIFY_STOP)
66		return;
67#endif
68
69	info.si_code = SEGV_MAPERR;
70
71	/*
72	 * We fault-in kernel-space virtual memory on-demand. The
73	 * 'reference' page table is init_mm.pgd.
74	 *
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,
78	 * nothing more.
79	 */
80#ifdef CONFIG_64BIT
81# define VMALLOC_FAULT_TARGET no_context
82#else
83# define VMALLOC_FAULT_TARGET vmalloc_fault
84#endif
85
86	if (unlikely(address >= VMALLOC_START && address <= VMALLOC_END))
87		goto VMALLOC_FAULT_TARGET;
88#ifdef MODULE_START
89	if (unlikely(address >= MODULE_START && address < MODULE_END))
90		goto VMALLOC_FAULT_TARGET;
91#endif
92
93	/*
94	 * If we're in an interrupt or have no user
95	 * context, we must not take the fault..
96	 */
97	if (in_atomic() || !mm)
98		goto bad_area_nosemaphore;
99
100	if (user_mode(regs))
101		flags |= FAULT_FLAG_USER;
102retry:
103	down_read(&mm->mmap_sem);
104	vma = find_vma(mm, address);
105	if (!vma)
106		goto bad_area;
107	if (vma->vm_start <= address)
108		goto good_area;
109	if (!(vma->vm_flags & VM_GROWSDOWN))
110		goto bad_area;
111	if (expand_stack(vma, address))
112		goto bad_area;
113/*
114 * Ok, we have a good vm_area for this memory access, so
115 * we can handle it..
116 */
117good_area:
118	info.si_code = SEGV_ACCERR;
119
120	if (write) {
121		if (!(vma->vm_flags & VM_WRITE))
122			goto bad_area;
123		flags |= FAULT_FLAG_WRITE;
124	} else {
125		if (cpu_has_rixi) {
126			if (address == regs->cp0_epc && !(vma->vm_flags & VM_EXEC)) {
127#if 0
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);
133#endif
134				goto bad_area;
135			}
136			if (!(vma->vm_flags & VM_READ)) {
137#if 0
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);
143#endif
144				goto bad_area;
145			}
146		} else {
147			if (!(vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC)))
148				goto bad_area;
149		}
150	}
151
152	/*
153	 * If for any reason at all we couldn't handle the fault,
154	 * make sure we exit gracefully rather than endlessly redo
155	 * the fault.
156	 */
157	fault = handle_mm_fault(mm, vma, address, flags);
158
159	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))
160		return;
161
162	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
163	if (unlikely(fault & VM_FAULT_ERROR)) {
164		if (fault & VM_FAULT_OOM)
165			goto out_of_memory;
166		else if (fault & VM_FAULT_SIGSEGV)
167			goto bad_area;
168		else if (fault & VM_FAULT_SIGBUS)
169			goto do_sigbus;
170		BUG();
171	}
172	if (flags & FAULT_FLAG_ALLOW_RETRY) {
173		if (fault & VM_FAULT_MAJOR) {
174			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
175						  regs, address);
176			tsk->maj_flt++;
177		} else {
178			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
179						  regs, address);
180			tsk->min_flt++;
181		}
182		if (fault & VM_FAULT_RETRY) {
183			flags &= ~FAULT_FLAG_ALLOW_RETRY;
184			flags |= FAULT_FLAG_TRIED;
185
186			/*
187			 * No need to up_read(&mm->mmap_sem) as we would
188			 * have already released it in __lock_page_or_retry
189			 * in mm/filemap.c.
190			 */
191
192			goto retry;
193		}
194	}
195
196	up_read(&mm->mmap_sem);
197	return;
198
199/*
200 * Something tried to access memory that isn't in our memory map..
201 * Fix it, but check if it's kernel or user first..
202 */
203bad_area:
204	up_read(&mm->mmap_sem);
205
206bad_area_nosemaphore:
207	/* User mode accesses just cause a SIGSEGV */
208	if (user_mode(regs)) {
209		tsk->thread.cp0_badvaddr = address;
210		tsk->thread.error_code = write;
211		if (show_unhandled_signals &&
212		    unhandled_signal(tsk, SIGSEGV) &&
213		    __ratelimit(&ratelimit_state)) {
214			pr_info("\ndo_page_fault(): sending SIGSEGV to %s for invalid %s %0*lx",
215				tsk->comm,
216				write ? "write access to" : "read access from",
217				field, address);
218			pr_info("epc = %0*lx in", field,
219				(unsigned long) regs->cp0_epc);
220			print_vma_addr(" ", regs->cp0_epc);
221			pr_info("ra  = %0*lx in", field,
222				(unsigned long) regs->regs[31]);
223			print_vma_addr(" ", regs->regs[31]);
224			pr_info("\n");
225		}
226		info.si_signo = SIGSEGV;
227		info.si_errno = 0;
228		/* info.si_code has been set above */
229		info.si_addr = (void __user *) address;
230		force_sig_info(SIGSEGV, &info, tsk);
231		return;
232	}
233
234no_context:
235	/* Are we prepared to handle this kernel fault?	 */
236	if (fixup_exception(regs)) {
237		current->thread.cp0_baduaddr = address;
238		return;
239	}
240
241	/*
242	 * Oops. The kernel tried to access some bad page. We'll have to
243	 * terminate things with extreme prejudice.
244	 */
245	bust_spinlocks(1);
246
247	printk(KERN_ALERT "CPU %d Unable to handle kernel paging request at "
248	       "virtual address %0*lx, epc == %0*lx, ra == %0*lx\n",
249	       raw_smp_processor_id(), field, address, field, regs->cp0_epc,
250	       field,  regs->regs[31]);
251	die("Oops", regs);
252
253out_of_memory:
254	/*
255	 * We ran out of memory, call the OOM killer, and return the userspace
256	 * (which will retry the fault, or kill us if we got oom-killed).
257	 */
258	up_read(&mm->mmap_sem);
259	if (!user_mode(regs))
260		goto no_context;
261	pagefault_out_of_memory();
262	return;
263
264do_sigbus:
265	up_read(&mm->mmap_sem);
266
267	/* Kernel mode? Handle exceptions or die */
268	if (!user_mode(regs))
269		goto no_context;
270	else
271	/*
272	 * Send a sigbus, regardless of whether we were in kernel
273	 * or user mode.
274	 */
275#if 0
276		printk("do_page_fault() #3: sending SIGBUS to %s for "
277		       "invalid %s\n%0*lx (epc == %0*lx, ra == %0*lx)\n",
278		       tsk->comm,
279		       write ? "write access to" : "read access from",
280		       field, address,
281		       field, (unsigned long) regs->cp0_epc,
282		       field, (unsigned long) regs->regs[31]);
283#endif
284	tsk->thread.cp0_badvaddr = address;
285	info.si_signo = SIGBUS;
286	info.si_errno = 0;
287	info.si_code = BUS_ADRERR;
288	info.si_addr = (void __user *) address;
289	force_sig_info(SIGBUS, &info, tsk);
290
291	return;
292#ifndef CONFIG_64BIT
293vmalloc_fault:
294	{
295		/*
296		 * Synchronize this task's top level page-table
297		 * with the 'reference' page table.
298		 *
299		 * Do _not_ use "tsk" here. We might be inside
300		 * an interrupt in the middle of a task switch..
301		 */
302		int offset = __pgd_offset(address);
303		pgd_t *pgd, *pgd_k;
304		pud_t *pud, *pud_k;
305		pmd_t *pmd, *pmd_k;
306		pte_t *pte_k;
307
308		pgd = (pgd_t *) pgd_current[raw_smp_processor_id()] + offset;
309		pgd_k = init_mm.pgd + offset;
310
311		if (!pgd_present(*pgd_k))
312			goto no_context;
313		set_pgd(pgd, *pgd_k);
314
315		pud = pud_offset(pgd, address);
316		pud_k = pud_offset(pgd_k, address);
317		if (!pud_present(*pud_k))
318			goto no_context;
319
320		pmd = pmd_offset(pud, address);
321		pmd_k = pmd_offset(pud_k, address);
322		if (!pmd_present(*pmd_k))
323			goto no_context;
324		set_pmd(pmd, *pmd_k);
325
326		pte_k = pte_offset_kernel(pmd_k, address);
327		if (!pte_present(*pte_k))
328			goto no_context;
329		return;
330	}
331#endif
332}
333
334asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
335	unsigned long write, unsigned long address)
336{
337	enum ctx_state prev_state;
338
339	prev_state = exception_enter();
340	__do_page_fault(regs, write, address);
341	exception_exit(prev_state);
342}
343