xref: /linux/arch/s390/mm/fault.c (revision b0148a98ec5151fec82064d95f11eb9efbc628ea)
1 /*
2  *  arch/s390/mm/fault.c
3  *
4  *  S390 version
5  *    Copyright (C) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
6  *    Author(s): Hartmut Penner (hp@de.ibm.com)
7  *               Ulrich Weigand (uweigand@de.ibm.com)
8  *
9  *  Derived from "arch/i386/mm/fault.c"
10  *    Copyright (C) 1995  Linus Torvalds
11  */
12 
13 #include <linux/signal.h>
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/errno.h>
17 #include <linux/string.h>
18 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/mman.h>
21 #include <linux/mm.h>
22 #include <linux/smp.h>
23 #include <linux/smp_lock.h>
24 #include <linux/init.h>
25 #include <linux/console.h>
26 #include <linux/module.h>
27 #include <linux/hardirq.h>
28 #include <linux/kprobes.h>
29 
30 #include <asm/system.h>
31 #include <asm/uaccess.h>
32 #include <asm/pgtable.h>
33 #include <asm/kdebug.h>
34 #include <asm/s390_ext.h>
35 
36 #ifndef CONFIG_64BIT
37 #define __FAIL_ADDR_MASK 0x7ffff000
38 #define __FIXUP_MASK 0x7fffffff
39 #define __SUBCODE_MASK 0x0200
40 #define __PF_RES_FIELD 0ULL
41 #else /* CONFIG_64BIT */
42 #define __FAIL_ADDR_MASK -4096L
43 #define __FIXUP_MASK ~0L
44 #define __SUBCODE_MASK 0x0600
45 #define __PF_RES_FIELD 0x8000000000000000ULL
46 #endif /* CONFIG_64BIT */
47 
48 #ifdef CONFIG_SYSCTL
49 extern int sysctl_userprocess_debug;
50 #endif
51 
52 extern void die(const char *,struct pt_regs *,long);
53 
54 #ifdef CONFIG_KPROBES
55 static ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
56 int register_page_fault_notifier(struct notifier_block *nb)
57 {
58 	return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
59 }
60 
61 int unregister_page_fault_notifier(struct notifier_block *nb)
62 {
63 	return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
64 }
65 
66 static inline int notify_page_fault(enum die_val val, const char *str,
67 			struct pt_regs *regs, long err, int trap, int sig)
68 {
69 	struct die_args args = {
70 		.regs = regs,
71 		.str = str,
72 		.err = err,
73 		.trapnr = trap,
74 		.signr = sig
75 	};
76 	return atomic_notifier_call_chain(&notify_page_fault_chain, val, &args);
77 }
78 #else
79 static inline int notify_page_fault(enum die_val val, const char *str,
80 			struct pt_regs *regs, long err, int trap, int sig)
81 {
82 	return NOTIFY_DONE;
83 }
84 #endif
85 
86 extern spinlock_t timerlist_lock;
87 
88 /*
89  * Unlock any spinlocks which will prevent us from getting the
90  * message out (timerlist_lock is acquired through the
91  * console unblank code)
92  */
93 void bust_spinlocks(int yes)
94 {
95 	if (yes) {
96 		oops_in_progress = 1;
97 	} else {
98 		int loglevel_save = console_loglevel;
99 		console_unblank();
100 		oops_in_progress = 0;
101 		/*
102 		 * OK, the message is on the console.  Now we call printk()
103 		 * without oops_in_progress set so that printk will give klogd
104 		 * a poke.  Hold onto your hats...
105 		 */
106 		console_loglevel = 15;
107 		printk(" ");
108 		console_loglevel = loglevel_save;
109 	}
110 }
111 
112 /*
113  * Check which address space is addressed by the access
114  * register in S390_lowcore.exc_access_id.
115  * Returns 1 for user space and 0 for kernel space.
116  */
117 static int __check_access_register(struct pt_regs *regs, int error_code)
118 {
119 	int areg = S390_lowcore.exc_access_id;
120 
121 	if (areg == 0)
122 		/* Access via access register 0 -> kernel address */
123 		return 0;
124 	save_access_regs(current->thread.acrs);
125 	if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
126 		/*
127 		 * access register contains 0 -> kernel address,
128 		 * access register contains 1 -> user space address
129 		 */
130 		return current->thread.acrs[areg];
131 
132 	/* Something unhealthy was done with the access registers... */
133 	die("page fault via unknown access register", regs, error_code);
134 	do_exit(SIGKILL);
135 	return 0;
136 }
137 
138 /*
139  * Check which address space the address belongs to.
140  * May return 1 or 2 for user space and 0 for kernel space.
141  * Returns 2 for user space in primary addressing mode with
142  * CONFIG_S390_EXEC_PROTECT on and kernel parameter noexec=on.
143  */
144 static inline int check_user_space(struct pt_regs *regs, int error_code)
145 {
146 	/*
147 	 * The lowest two bits of S390_lowcore.trans_exc_code indicate
148 	 * which paging table was used:
149 	 *   0: Primary Segment Table Descriptor
150 	 *   1: STD determined via access register
151 	 *   2: Secondary Segment Table Descriptor
152 	 *   3: Home Segment Table Descriptor
153 	 */
154 	int descriptor = S390_lowcore.trans_exc_code & 3;
155 	if (unlikely(descriptor == 1))
156 		return __check_access_register(regs, error_code);
157 	if (descriptor == 2)
158 		return current->thread.mm_segment.ar4;
159 	return ((descriptor != 0) ^ (switch_amode)) << s390_noexec;
160 }
161 
162 /*
163  * Send SIGSEGV to task.  This is an external routine
164  * to keep the stack usage of do_page_fault small.
165  */
166 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
167 		       int si_code, unsigned long address)
168 {
169 	struct siginfo si;
170 
171 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
172 #if defined(CONFIG_SYSCTL)
173 	if (sysctl_userprocess_debug)
174 #endif
175 	{
176 		printk("User process fault: interruption code 0x%lX\n",
177 		       error_code);
178 		printk("failing address: %lX\n", address);
179 		show_regs(regs);
180 	}
181 #endif
182 	si.si_signo = SIGSEGV;
183 	si.si_code = si_code;
184 	si.si_addr = (void __user *) address;
185 	force_sig_info(SIGSEGV, &si, current);
186 }
187 
188 #ifdef CONFIG_S390_EXEC_PROTECT
189 extern long sys_sigreturn(struct pt_regs *regs);
190 extern long sys_rt_sigreturn(struct pt_regs *regs);
191 extern long sys32_sigreturn(struct pt_regs *regs);
192 extern long sys32_rt_sigreturn(struct pt_regs *regs);
193 
194 static inline void do_sigreturn(struct mm_struct *mm, struct pt_regs *regs,
195 				int rt)
196 {
197 	up_read(&mm->mmap_sem);
198 	clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
199 #ifdef CONFIG_COMPAT
200 	if (test_tsk_thread_flag(current, TIF_31BIT)) {
201 		if (rt)
202 			sys32_rt_sigreturn(regs);
203 		else
204 			sys32_sigreturn(regs);
205 		return;
206 	}
207 #endif /* CONFIG_COMPAT */
208 	if (rt)
209 		sys_rt_sigreturn(regs);
210 	else
211 		sys_sigreturn(regs);
212 	return;
213 }
214 
215 static int signal_return(struct mm_struct *mm, struct pt_regs *regs,
216 			 unsigned long address, unsigned long error_code)
217 {
218 	pgd_t *pgd;
219 	pmd_t *pmd;
220 	pte_t *pte;
221 	u16 *instruction;
222 	unsigned long pfn, uaddr = regs->psw.addr;
223 
224 	spin_lock(&mm->page_table_lock);
225 	pgd = pgd_offset(mm, uaddr);
226 	if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
227 		goto out_fault;
228 	pmd = pmd_offset(pgd, uaddr);
229 	if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
230 		goto out_fault;
231 	pte = pte_offset_map(pmd_offset(pgd_offset(mm, uaddr), uaddr), uaddr);
232 	if (!pte || !pte_present(*pte))
233 		goto out_fault;
234 	pfn = pte_pfn(*pte);
235 	if (!pfn_valid(pfn))
236 		goto out_fault;
237 	spin_unlock(&mm->page_table_lock);
238 
239 	instruction = (u16 *) ((pfn << PAGE_SHIFT) + (uaddr & (PAGE_SIZE-1)));
240 	if (*instruction == 0x0a77)
241 		do_sigreturn(mm, regs, 0);
242 	else if (*instruction == 0x0aad)
243 		do_sigreturn(mm, regs, 1);
244 	else {
245 		printk("- XXX - do_exception: task = %s, primary, NO EXEC "
246 		       "-> SIGSEGV\n", current->comm);
247 		up_read(&mm->mmap_sem);
248 		current->thread.prot_addr = address;
249 		current->thread.trap_no = error_code;
250 		do_sigsegv(regs, error_code, SEGV_MAPERR, address);
251 	}
252 	return 0;
253 out_fault:
254 	spin_unlock(&mm->page_table_lock);
255 	return -EFAULT;
256 }
257 #endif /* CONFIG_S390_EXEC_PROTECT */
258 
259 /*
260  * This routine handles page faults.  It determines the address,
261  * and the problem, and then passes it off to one of the appropriate
262  * routines.
263  *
264  * error_code:
265  *   04       Protection           ->  Write-Protection  (suprression)
266  *   10       Segment translation  ->  Not present       (nullification)
267  *   11       Page translation     ->  Not present       (nullification)
268  *   3b       Region third trans.  ->  Not present       (nullification)
269  */
270 static inline void __kprobes
271 do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection)
272 {
273         struct task_struct *tsk;
274         struct mm_struct *mm;
275         struct vm_area_struct * vma;
276         unsigned long address;
277 	int user_address;
278 	const struct exception_table_entry *fixup;
279 	int si_code = SEGV_MAPERR;
280 
281         tsk = current;
282         mm = tsk->mm;
283 
284 	if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14,
285 					SIGSEGV) == NOTIFY_STOP)
286 		return;
287 
288 	/*
289          * Check for low-address protection.  This needs to be treated
290 	 * as a special case because the translation exception code
291 	 * field is not guaranteed to contain valid data in this case.
292 	 */
293 	if (is_protection && !(S390_lowcore.trans_exc_code & 4)) {
294 
295 		/* Low-address protection hit in kernel mode means
296 		   NULL pointer write access in kernel mode.  */
297  		if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
298 			address = 0;
299 			user_address = 0;
300 			goto no_context;
301 		}
302 
303 		/* Low-address protection hit in user mode 'cannot happen'.  */
304 		die ("Low-address protection", regs, error_code);
305         	do_exit(SIGKILL);
306 	}
307 
308         /*
309          * get the failing address
310          * more specific the segment and page table portion of
311          * the address
312          */
313         address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
314 	user_address = check_user_space(regs, error_code);
315 
316 	/*
317 	 * Verify that the fault happened in user space, that
318 	 * we are not in an interrupt and that there is a
319 	 * user context.
320 	 */
321         if (user_address == 0 || in_atomic() || !mm)
322                 goto no_context;
323 
324 	/*
325 	 * When we get here, the fault happened in the current
326 	 * task's user address space, so we can switch on the
327 	 * interrupts again and then search the VMAs
328 	 */
329 	local_irq_enable();
330 
331         down_read(&mm->mmap_sem);
332 
333         vma = find_vma(mm, address);
334         if (!vma)
335                 goto bad_area;
336 
337 #ifdef CONFIG_S390_EXEC_PROTECT
338 	if (unlikely((user_address == 2) && !(vma->vm_flags & VM_EXEC)))
339 		if (!signal_return(mm, regs, address, error_code))
340 			/*
341 			 * signal_return() has done an up_read(&mm->mmap_sem)
342 			 * if it returns 0.
343 			 */
344 			return;
345 #endif
346 
347         if (vma->vm_start <= address)
348                 goto good_area;
349         if (!(vma->vm_flags & VM_GROWSDOWN))
350                 goto bad_area;
351         if (expand_stack(vma, address))
352                 goto bad_area;
353 /*
354  * Ok, we have a good vm_area for this memory access, so
355  * we can handle it..
356  */
357 good_area:
358 	si_code = SEGV_ACCERR;
359 	if (!is_protection) {
360 		/* page not present, check vm flags */
361 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
362 			goto bad_area;
363 	} else {
364 		if (!(vma->vm_flags & VM_WRITE))
365 			goto bad_area;
366 	}
367 
368 survive:
369 	/*
370 	 * If for any reason at all we couldn't handle the fault,
371 	 * make sure we exit gracefully rather than endlessly redo
372 	 * the fault.
373 	 */
374 	switch (handle_mm_fault(mm, vma, address, is_protection)) {
375 	case VM_FAULT_MINOR:
376 		tsk->min_flt++;
377 		break;
378 	case VM_FAULT_MAJOR:
379 		tsk->maj_flt++;
380 		break;
381 	case VM_FAULT_SIGBUS:
382 		goto do_sigbus;
383 	case VM_FAULT_OOM:
384 		goto out_of_memory;
385 	default:
386 		BUG();
387 	}
388 
389         up_read(&mm->mmap_sem);
390 	/*
391 	 * The instruction that caused the program check will
392 	 * be repeated. Don't signal single step via SIGTRAP.
393 	 */
394 	clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
395         return;
396 
397 /*
398  * Something tried to access memory that isn't in our memory map..
399  * Fix it, but check if it's kernel or user first..
400  */
401 bad_area:
402         up_read(&mm->mmap_sem);
403 
404         /* User mode accesses just cause a SIGSEGV */
405         if (regs->psw.mask & PSW_MASK_PSTATE) {
406                 tsk->thread.prot_addr = address;
407                 tsk->thread.trap_no = error_code;
408 		do_sigsegv(regs, error_code, si_code, address);
409                 return;
410 	}
411 
412 no_context:
413         /* Are we prepared to handle this kernel fault?  */
414 	fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
415 	if (fixup) {
416 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
417                 return;
418         }
419 
420 /*
421  * Oops. The kernel tried to access some bad page. We'll have to
422  * terminate things with extreme prejudice.
423  */
424         if (user_address == 0)
425                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
426         	       " at virtual kernel address %p\n", (void *)address);
427         else
428                 printk(KERN_ALERT "Unable to handle kernel paging request"
429 		       " at virtual user address %p\n", (void *)address);
430 
431         die("Oops", regs, error_code);
432         do_exit(SIGKILL);
433 
434 
435 /*
436  * We ran out of memory, or some other thing happened to us that made
437  * us unable to handle the page fault gracefully.
438 */
439 out_of_memory:
440 	up_read(&mm->mmap_sem);
441 	if (is_init(tsk)) {
442 		yield();
443 		down_read(&mm->mmap_sem);
444 		goto survive;
445 	}
446 	printk("VM: killing process %s\n", tsk->comm);
447 	if (regs->psw.mask & PSW_MASK_PSTATE)
448 		do_exit(SIGKILL);
449 	goto no_context;
450 
451 do_sigbus:
452 	up_read(&mm->mmap_sem);
453 
454 	/*
455 	 * Send a sigbus, regardless of whether we were in kernel
456 	 * or user mode.
457 	 */
458         tsk->thread.prot_addr = address;
459         tsk->thread.trap_no = error_code;
460 	force_sig(SIGBUS, tsk);
461 
462 	/* Kernel mode? Handle exceptions or die */
463 	if (!(regs->psw.mask & PSW_MASK_PSTATE))
464 		goto no_context;
465 }
466 
467 void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
468 {
469 	regs->psw.addr -= (error_code >> 16);
470 	do_exception(regs, 4, 1);
471 }
472 
473 void do_dat_exception(struct pt_regs *regs, unsigned long error_code)
474 {
475 	do_exception(regs, error_code & 0xff, 0);
476 }
477 
478 #ifdef CONFIG_PFAULT
479 /*
480  * 'pfault' pseudo page faults routines.
481  */
482 static ext_int_info_t ext_int_pfault;
483 static int pfault_disable = 0;
484 
485 static int __init nopfault(char *str)
486 {
487 	pfault_disable = 1;
488 	return 1;
489 }
490 
491 __setup("nopfault", nopfault);
492 
493 typedef struct {
494 	__u16 refdiagc;
495 	__u16 reffcode;
496 	__u16 refdwlen;
497 	__u16 refversn;
498 	__u64 refgaddr;
499 	__u64 refselmk;
500 	__u64 refcmpmk;
501 	__u64 reserved;
502 } __attribute__ ((packed)) pfault_refbk_t;
503 
504 int pfault_init(void)
505 {
506 	pfault_refbk_t refbk =
507 		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
508 		  __PF_RES_FIELD };
509         int rc;
510 
511 	if (!MACHINE_IS_VM || pfault_disable)
512 		return -1;
513 	asm volatile(
514 		"	diag	%1,%0,0x258\n"
515 		"0:	j	2f\n"
516 		"1:	la	%0,8\n"
517 		"2:\n"
518 		EX_TABLE(0b,1b)
519 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
520         __ctl_set_bit(0, 9);
521         return rc;
522 }
523 
524 void pfault_fini(void)
525 {
526 	pfault_refbk_t refbk =
527 	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
528 
529 	if (!MACHINE_IS_VM || pfault_disable)
530 		return;
531 	__ctl_clear_bit(0,9);
532 	asm volatile(
533 		"	diag	%0,0,0x258\n"
534 		"0:\n"
535 		EX_TABLE(0b,0b)
536 		: : "a" (&refbk), "m" (refbk) : "cc");
537 }
538 
539 static void pfault_interrupt(__u16 error_code)
540 {
541 	struct task_struct *tsk;
542 	__u16 subcode;
543 
544 	/*
545 	 * Get the external interruption subcode & pfault
546 	 * initial/completion signal bit. VM stores this
547 	 * in the 'cpu address' field associated with the
548          * external interrupt.
549 	 */
550 	subcode = S390_lowcore.cpu_addr;
551 	if ((subcode & 0xff00) != __SUBCODE_MASK)
552 		return;
553 
554 	/*
555 	 * Get the token (= address of the task structure of the affected task).
556 	 */
557 	tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
558 
559 	if (subcode & 0x0080) {
560 		/* signal bit is set -> a page has been swapped in by VM */
561 		if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
562 			/* Initial interrupt was faster than the completion
563 			 * interrupt. pfault_wait is valid. Set pfault_wait
564 			 * back to zero and wake up the process. This can
565 			 * safely be done because the task is still sleeping
566 			 * and can't produce new pfaults. */
567 			tsk->thread.pfault_wait = 0;
568 			wake_up_process(tsk);
569 			put_task_struct(tsk);
570 		}
571 	} else {
572 		/* signal bit not set -> a real page is missing. */
573 		get_task_struct(tsk);
574 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
575 		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
576 			/* Completion interrupt was faster than the initial
577 			 * interrupt (swapped in a -1 for pfault_wait). Set
578 			 * pfault_wait back to zero and exit. This can be
579 			 * done safely because tsk is running in kernel
580 			 * mode and can't produce new pfaults. */
581 			tsk->thread.pfault_wait = 0;
582 			set_task_state(tsk, TASK_RUNNING);
583 			put_task_struct(tsk);
584 		} else
585 			set_tsk_need_resched(tsk);
586 	}
587 }
588 
589 void __init pfault_irq_init(void)
590 {
591 	if (!MACHINE_IS_VM)
592 		return;
593 
594 	/*
595 	 * Try to get pfault pseudo page faults going.
596 	 */
597 	if (register_early_external_interrupt(0x2603, pfault_interrupt,
598 					      &ext_int_pfault) != 0)
599 		panic("Couldn't request external interrupt 0x2603");
600 
601 	if (pfault_init() == 0)
602 		return;
603 
604 	/* Tough luck, no pfault. */
605 	pfault_disable = 1;
606 	unregister_early_external_interrupt(0x2603, pfault_interrupt,
607 					    &ext_int_pfault);
608 }
609 #endif
610