xref: /openbmc/linux/arch/sh/mm/fault.c (revision a44e4f3a)
1 /*
2  * Page fault handler for SH with an MMU.
3  *
4  *  Copyright (C) 1999  Niibe Yutaka
5  *  Copyright (C) 2003 - 2012  Paul Mundt
6  *
7  *  Based on linux/arch/i386/mm/fault.c:
8  *   Copyright (C) 1995  Linus Torvalds
9  *
10  * This file is subject to the terms and conditions of the GNU General Public
11  * License.  See the file "COPYING" in the main directory of this archive
12  * for more details.
13  */
14 #include <linux/kernel.h>
15 #include <linux/mm.h>
16 #include <linux/sched/signal.h>
17 #include <linux/hardirq.h>
18 #include <linux/kprobes.h>
19 #include <linux/perf_event.h>
20 #include <linux/kdebug.h>
21 #include <linux/uaccess.h>
22 #include <asm/io_trapped.h>
23 #include <asm/mmu_context.h>
24 #include <asm/tlbflush.h>
25 #include <asm/traps.h>
26 
27 static void
28 force_sig_info_fault(int si_signo, int si_code, unsigned long address)
29 {
30 	force_sig_fault(si_signo, si_code, (void __user *)address);
31 }
32 
33 /*
34  * This is useful to dump out the page tables associated with
35  * 'addr' in mm 'mm'.
36  */
37 static void show_pte(struct mm_struct *mm, unsigned long addr)
38 {
39 	pgd_t *pgd;
40 
41 	if (mm) {
42 		pgd = mm->pgd;
43 	} else {
44 		pgd = get_TTB();
45 
46 		if (unlikely(!pgd))
47 			pgd = swapper_pg_dir;
48 	}
49 
50 	printk(KERN_ALERT "pgd = %p\n", pgd);
51 	pgd += pgd_index(addr);
52 	printk(KERN_ALERT "[%08lx] *pgd=%0*Lx", addr,
53 	       (u32)(sizeof(*pgd) * 2), (u64)pgd_val(*pgd));
54 
55 	do {
56 		pud_t *pud;
57 		pmd_t *pmd;
58 		pte_t *pte;
59 
60 		if (pgd_none(*pgd))
61 			break;
62 
63 		if (pgd_bad(*pgd)) {
64 			printk("(bad)");
65 			break;
66 		}
67 
68 		pud = pud_offset(pgd, addr);
69 		if (PTRS_PER_PUD != 1)
70 			printk(", *pud=%0*Lx", (u32)(sizeof(*pud) * 2),
71 			       (u64)pud_val(*pud));
72 
73 		if (pud_none(*pud))
74 			break;
75 
76 		if (pud_bad(*pud)) {
77 			printk("(bad)");
78 			break;
79 		}
80 
81 		pmd = pmd_offset(pud, addr);
82 		if (PTRS_PER_PMD != 1)
83 			printk(", *pmd=%0*Lx", (u32)(sizeof(*pmd) * 2),
84 			       (u64)pmd_val(*pmd));
85 
86 		if (pmd_none(*pmd))
87 			break;
88 
89 		if (pmd_bad(*pmd)) {
90 			printk("(bad)");
91 			break;
92 		}
93 
94 		/* We must not map this if we have highmem enabled */
95 		if (PageHighMem(pfn_to_page(pmd_val(*pmd) >> PAGE_SHIFT)))
96 			break;
97 
98 		pte = pte_offset_kernel(pmd, addr);
99 		printk(", *pte=%0*Lx", (u32)(sizeof(*pte) * 2),
100 		       (u64)pte_val(*pte));
101 	} while (0);
102 
103 	printk("\n");
104 }
105 
106 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
107 {
108 	unsigned index = pgd_index(address);
109 	pgd_t *pgd_k;
110 	pud_t *pud, *pud_k;
111 	pmd_t *pmd, *pmd_k;
112 
113 	pgd += index;
114 	pgd_k = init_mm.pgd + index;
115 
116 	if (!pgd_present(*pgd_k))
117 		return NULL;
118 
119 	pud = pud_offset(pgd, address);
120 	pud_k = pud_offset(pgd_k, address);
121 	if (!pud_present(*pud_k))
122 		return NULL;
123 
124 	if (!pud_present(*pud))
125 	    set_pud(pud, *pud_k);
126 
127 	pmd = pmd_offset(pud, address);
128 	pmd_k = pmd_offset(pud_k, address);
129 	if (!pmd_present(*pmd_k))
130 		return NULL;
131 
132 	if (!pmd_present(*pmd))
133 		set_pmd(pmd, *pmd_k);
134 	else {
135 		/*
136 		 * The page tables are fully synchronised so there must
137 		 * be another reason for the fault. Return NULL here to
138 		 * signal that we have not taken care of the fault.
139 		 */
140 		BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
141 		return NULL;
142 	}
143 
144 	return pmd_k;
145 }
146 
147 #ifdef CONFIG_SH_STORE_QUEUES
148 #define __FAULT_ADDR_LIMIT	P3_ADDR_MAX
149 #else
150 #define __FAULT_ADDR_LIMIT	VMALLOC_END
151 #endif
152 
153 /*
154  * Handle a fault on the vmalloc or module mapping area
155  */
156 static noinline int vmalloc_fault(unsigned long address)
157 {
158 	pgd_t *pgd_k;
159 	pmd_t *pmd_k;
160 	pte_t *pte_k;
161 
162 	/* Make sure we are in vmalloc/module/P3 area: */
163 	if (!(address >= VMALLOC_START && address < __FAULT_ADDR_LIMIT))
164 		return -1;
165 
166 	/*
167 	 * Synchronize this task's top level page-table
168 	 * with the 'reference' page table.
169 	 *
170 	 * Do _not_ use "current" here. We might be inside
171 	 * an interrupt in the middle of a task switch..
172 	 */
173 	pgd_k = get_TTB();
174 	pmd_k = vmalloc_sync_one(pgd_k, address);
175 	if (!pmd_k)
176 		return -1;
177 
178 	pte_k = pte_offset_kernel(pmd_k, address);
179 	if (!pte_present(*pte_k))
180 		return -1;
181 
182 	return 0;
183 }
184 
185 static void
186 show_fault_oops(struct pt_regs *regs, unsigned long address)
187 {
188 	if (!oops_may_print())
189 		return;
190 
191 	printk(KERN_ALERT "BUG: unable to handle kernel ");
192 	if (address < PAGE_SIZE)
193 		printk(KERN_CONT "NULL pointer dereference");
194 	else
195 		printk(KERN_CONT "paging request");
196 
197 	printk(KERN_CONT " at %08lx\n", address);
198 	printk(KERN_ALERT "PC:");
199 	printk_address(regs->pc, 1);
200 
201 	show_pte(NULL, address);
202 }
203 
204 static noinline void
205 no_context(struct pt_regs *regs, unsigned long error_code,
206 	   unsigned long address)
207 {
208 	/* Are we prepared to handle this kernel fault?  */
209 	if (fixup_exception(regs))
210 		return;
211 
212 	if (handle_trapped_io(regs, address))
213 		return;
214 
215 	/*
216 	 * Oops. The kernel tried to access some bad page. We'll have to
217 	 * terminate things with extreme prejudice.
218 	 */
219 	bust_spinlocks(1);
220 
221 	show_fault_oops(regs, address);
222 
223 	die("Oops", regs, error_code);
224 	bust_spinlocks(0);
225 	do_exit(SIGKILL);
226 }
227 
228 static void
229 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
230 		       unsigned long address, int si_code)
231 {
232 	/* User mode accesses just cause a SIGSEGV */
233 	if (user_mode(regs)) {
234 		/*
235 		 * It's possible to have interrupts off here:
236 		 */
237 		local_irq_enable();
238 
239 		force_sig_info_fault(SIGSEGV, si_code, address);
240 
241 		return;
242 	}
243 
244 	no_context(regs, error_code, address);
245 }
246 
247 static noinline void
248 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
249 		     unsigned long address)
250 {
251 	__bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
252 }
253 
254 static void
255 __bad_area(struct pt_regs *regs, unsigned long error_code,
256 	   unsigned long address, int si_code)
257 {
258 	struct mm_struct *mm = current->mm;
259 
260 	/*
261 	 * Something tried to access memory that isn't in our memory map..
262 	 * Fix it, but check if it's kernel or user first..
263 	 */
264 	up_read(&mm->mmap_sem);
265 
266 	__bad_area_nosemaphore(regs, error_code, address, si_code);
267 }
268 
269 static noinline void
270 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
271 {
272 	__bad_area(regs, error_code, address, SEGV_MAPERR);
273 }
274 
275 static noinline void
276 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
277 		      unsigned long address)
278 {
279 	__bad_area(regs, error_code, address, SEGV_ACCERR);
280 }
281 
282 static void
283 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address)
284 {
285 	struct task_struct *tsk = current;
286 	struct mm_struct *mm = tsk->mm;
287 
288 	up_read(&mm->mmap_sem);
289 
290 	/* Kernel mode? Handle exceptions or die: */
291 	if (!user_mode(regs))
292 		no_context(regs, error_code, address);
293 
294 	force_sig_info_fault(SIGBUS, BUS_ADRERR, address);
295 }
296 
297 static noinline int
298 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
299 	       unsigned long address, vm_fault_t fault)
300 {
301 	/*
302 	 * Pagefault was interrupted by SIGKILL. We have no reason to
303 	 * continue pagefault.
304 	 */
305 	if (fatal_signal_pending(current)) {
306 		if (!(fault & VM_FAULT_RETRY))
307 			up_read(&current->mm->mmap_sem);
308 		if (!user_mode(regs))
309 			no_context(regs, error_code, address);
310 		return 1;
311 	}
312 
313 	if (!(fault & VM_FAULT_ERROR))
314 		return 0;
315 
316 	if (fault & VM_FAULT_OOM) {
317 		/* Kernel mode? Handle exceptions or die: */
318 		if (!user_mode(regs)) {
319 			up_read(&current->mm->mmap_sem);
320 			no_context(regs, error_code, address);
321 			return 1;
322 		}
323 		up_read(&current->mm->mmap_sem);
324 
325 		/*
326 		 * We ran out of memory, call the OOM killer, and return the
327 		 * userspace (which will retry the fault, or kill us if we got
328 		 * oom-killed):
329 		 */
330 		pagefault_out_of_memory();
331 	} else {
332 		if (fault & VM_FAULT_SIGBUS)
333 			do_sigbus(regs, error_code, address);
334 		else if (fault & VM_FAULT_SIGSEGV)
335 			bad_area(regs, error_code, address);
336 		else
337 			BUG();
338 	}
339 
340 	return 1;
341 }
342 
343 static inline int access_error(int error_code, struct vm_area_struct *vma)
344 {
345 	if (error_code & FAULT_CODE_WRITE) {
346 		/* write, present and write, not present: */
347 		if (unlikely(!(vma->vm_flags & VM_WRITE)))
348 			return 1;
349 		return 0;
350 	}
351 
352 	/* ITLB miss on NX page */
353 	if (unlikely((error_code & FAULT_CODE_ITLB) &&
354 		     !(vma->vm_flags & VM_EXEC)))
355 		return 1;
356 
357 	/* read, not present: */
358 	if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
359 		return 1;
360 
361 	return 0;
362 }
363 
364 static int fault_in_kernel_space(unsigned long address)
365 {
366 	return address >= TASK_SIZE;
367 }
368 
369 /*
370  * This routine handles page faults.  It determines the address,
371  * and the problem, and then passes it off to one of the appropriate
372  * routines.
373  */
374 asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
375 					unsigned long error_code,
376 					unsigned long address)
377 {
378 	unsigned long vec;
379 	struct task_struct *tsk;
380 	struct mm_struct *mm;
381 	struct vm_area_struct * vma;
382 	vm_fault_t fault;
383 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
384 
385 	tsk = current;
386 	mm = tsk->mm;
387 	vec = lookup_exception_vector();
388 
389 	/*
390 	 * We fault-in kernel-space virtual memory on-demand. The
391 	 * 'reference' page table is init_mm.pgd.
392 	 *
393 	 * NOTE! We MUST NOT take any locks for this case. We may
394 	 * be in an interrupt or a critical region, and should
395 	 * only copy the information from the master page table,
396 	 * nothing more.
397 	 */
398 	if (unlikely(fault_in_kernel_space(address))) {
399 		if (vmalloc_fault(address) >= 0)
400 			return;
401 		if (kprobe_page_fault(regs, vec))
402 			return;
403 
404 		bad_area_nosemaphore(regs, error_code, address);
405 		return;
406 	}
407 
408 	if (unlikely(kprobe_page_fault(regs, vec)))
409 		return;
410 
411 	/* Only enable interrupts if they were on before the fault */
412 	if ((regs->sr & SR_IMASK) != SR_IMASK)
413 		local_irq_enable();
414 
415 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
416 
417 	/*
418 	 * If we're in an interrupt, have no user context or are running
419 	 * with pagefaults disabled then we must not take the fault:
420 	 */
421 	if (unlikely(faulthandler_disabled() || !mm)) {
422 		bad_area_nosemaphore(regs, error_code, address);
423 		return;
424 	}
425 
426 retry:
427 	down_read(&mm->mmap_sem);
428 
429 	vma = find_vma(mm, address);
430 	if (unlikely(!vma)) {
431 		bad_area(regs, error_code, address);
432 		return;
433 	}
434 	if (likely(vma->vm_start <= address))
435 		goto good_area;
436 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
437 		bad_area(regs, error_code, address);
438 		return;
439 	}
440 	if (unlikely(expand_stack(vma, address))) {
441 		bad_area(regs, error_code, address);
442 		return;
443 	}
444 
445 	/*
446 	 * Ok, we have a good vm_area for this memory access, so
447 	 * we can handle it..
448 	 */
449 good_area:
450 	if (unlikely(access_error(error_code, vma))) {
451 		bad_area_access_error(regs, error_code, address);
452 		return;
453 	}
454 
455 	set_thread_fault_code(error_code);
456 
457 	if (user_mode(regs))
458 		flags |= FAULT_FLAG_USER;
459 	if (error_code & FAULT_CODE_WRITE)
460 		flags |= FAULT_FLAG_WRITE;
461 
462 	/*
463 	 * If for any reason at all we couldn't handle the fault,
464 	 * make sure we exit gracefully rather than endlessly redo
465 	 * the fault.
466 	 */
467 	fault = handle_mm_fault(vma, address, flags);
468 
469 	if (unlikely(fault & (VM_FAULT_RETRY | VM_FAULT_ERROR)))
470 		if (mm_fault_error(regs, error_code, address, fault))
471 			return;
472 
473 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
474 		if (fault & VM_FAULT_MAJOR) {
475 			tsk->maj_flt++;
476 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
477 				      regs, address);
478 		} else {
479 			tsk->min_flt++;
480 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
481 				      regs, address);
482 		}
483 		if (fault & VM_FAULT_RETRY) {
484 			flags &= ~FAULT_FLAG_ALLOW_RETRY;
485 			flags |= FAULT_FLAG_TRIED;
486 
487 			/*
488 			 * No need to up_read(&mm->mmap_sem) as we would
489 			 * have already released it in __lock_page_or_retry
490 			 * in mm/filemap.c.
491 			 */
492 			goto retry;
493 		}
494 	}
495 
496 	up_read(&mm->mmap_sem);
497 }
498