xref: /openbmc/linux/arch/s390/mm/fault.c (revision f42b3800)
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/kdebug.h>
24 #include <linux/smp_lock.h>
25 #include <linux/init.h>
26 #include <linux/console.h>
27 #include <linux/module.h>
28 #include <linux/hardirq.h>
29 #include <linux/kprobes.h>
30 #include <linux/uaccess.h>
31 #include <asm/system.h>
32 #include <asm/pgtable.h>
33 #include <asm/s390_ext.h>
34 #include <asm/mmu_context.h>
35 #include "../kernel/entry.h"
36 
37 #ifndef CONFIG_64BIT
38 #define __FAIL_ADDR_MASK 0x7ffff000
39 #define __FIXUP_MASK 0x7fffffff
40 #define __SUBCODE_MASK 0x0200
41 #define __PF_RES_FIELD 0ULL
42 #else /* CONFIG_64BIT */
43 #define __FAIL_ADDR_MASK -4096L
44 #define __FIXUP_MASK ~0L
45 #define __SUBCODE_MASK 0x0600
46 #define __PF_RES_FIELD 0x8000000000000000ULL
47 #endif /* CONFIG_64BIT */
48 
49 #ifdef CONFIG_SYSCTL
50 extern int sysctl_userprocess_debug;
51 #endif
52 
53 #ifdef CONFIG_KPROBES
54 static inline int notify_page_fault(struct pt_regs *regs, long err)
55 {
56 	int ret = 0;
57 
58 	/* kprobe_running() needs smp_processor_id() */
59 	if (!user_mode(regs)) {
60 		preempt_disable();
61 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
62 			ret = 1;
63 		preempt_enable();
64 	}
65 
66 	return ret;
67 }
68 #else
69 static inline int notify_page_fault(struct pt_regs *regs, long err)
70 {
71 	return 0;
72 }
73 #endif
74 
75 
76 /*
77  * Unlock any spinlocks which will prevent us from getting the
78  * message out.
79  */
80 void bust_spinlocks(int yes)
81 {
82 	if (yes) {
83 		oops_in_progress = 1;
84 	} else {
85 		int loglevel_save = console_loglevel;
86 		console_unblank();
87 		oops_in_progress = 0;
88 		/*
89 		 * OK, the message is on the console.  Now we call printk()
90 		 * without oops_in_progress set so that printk will give klogd
91 		 * a poke.  Hold onto your hats...
92 		 */
93 		console_loglevel = 15;
94 		printk(" ");
95 		console_loglevel = loglevel_save;
96 	}
97 }
98 
99 /*
100  * Returns the address space associated with the fault.
101  * Returns 0 for kernel space, 1 for user space and
102  * 2 for code execution in user space with noexec=on.
103  */
104 static inline int check_space(struct task_struct *tsk)
105 {
106 	/*
107 	 * The lowest two bits of S390_lowcore.trans_exc_code
108 	 * indicate which paging table was used.
109 	 */
110 	int desc = S390_lowcore.trans_exc_code & 3;
111 
112 	if (desc == 3)	/* Home Segment Table Descriptor */
113 		return switch_amode == 0;
114 	if (desc == 2)	/* Secondary Segment Table Descriptor */
115 		return tsk->thread.mm_segment.ar4;
116 #ifdef CONFIG_S390_SWITCH_AMODE
117 	if (unlikely(desc == 1)) { /* STD determined via access register */
118 		/* %a0 always indicates primary space. */
119 		if (S390_lowcore.exc_access_id != 0) {
120 			save_access_regs(tsk->thread.acrs);
121 			/*
122 			 * An alet of 0 indicates primary space.
123 			 * An alet of 1 indicates secondary space.
124 			 * Any other alet values generate an
125 			 * alen-translation exception.
126 			 */
127 			if (tsk->thread.acrs[S390_lowcore.exc_access_id])
128 				return tsk->thread.mm_segment.ar4;
129 		}
130 	}
131 #endif
132 	/* Primary Segment Table Descriptor */
133 	return switch_amode << s390_noexec;
134 }
135 
136 /*
137  * Send SIGSEGV to task.  This is an external routine
138  * to keep the stack usage of do_page_fault small.
139  */
140 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
141 		       int si_code, unsigned long address)
142 {
143 	struct siginfo si;
144 
145 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
146 #if defined(CONFIG_SYSCTL)
147 	if (sysctl_userprocess_debug)
148 #endif
149 	{
150 		printk("User process fault: interruption code 0x%lX\n",
151 		       error_code);
152 		printk("failing address: %lX\n", address);
153 		show_regs(regs);
154 	}
155 #endif
156 	si.si_signo = SIGSEGV;
157 	si.si_code = si_code;
158 	si.si_addr = (void __user *) address;
159 	force_sig_info(SIGSEGV, &si, current);
160 }
161 
162 static void do_no_context(struct pt_regs *regs, unsigned long error_code,
163 			  unsigned long address)
164 {
165 	const struct exception_table_entry *fixup;
166 
167 	/* Are we prepared to handle this kernel fault?  */
168 	fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
169 	if (fixup) {
170 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
171 		return;
172 	}
173 
174 	/*
175 	 * Oops. The kernel tried to access some bad page. We'll have to
176 	 * terminate things with extreme prejudice.
177 	 */
178 	if (check_space(current) == 0)
179 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
180 		       " at virtual kernel address %p\n", (void *)address);
181 	else
182 		printk(KERN_ALERT "Unable to handle kernel paging request"
183 		       " at virtual user address %p\n", (void *)address);
184 
185 	die("Oops", regs, error_code);
186 	do_exit(SIGKILL);
187 }
188 
189 static void do_low_address(struct pt_regs *regs, unsigned long error_code)
190 {
191 	/* Low-address protection hit in kernel mode means
192 	   NULL pointer write access in kernel mode.  */
193 	if (regs->psw.mask & PSW_MASK_PSTATE) {
194 		/* Low-address protection hit in user mode 'cannot happen'. */
195 		die ("Low-address protection", regs, error_code);
196 		do_exit(SIGKILL);
197 	}
198 
199 	do_no_context(regs, error_code, 0);
200 }
201 
202 /*
203  * We ran out of memory, or some other thing happened to us that made
204  * us unable to handle the page fault gracefully.
205  */
206 static int do_out_of_memory(struct pt_regs *regs, unsigned long error_code,
207 			    unsigned long address)
208 {
209 	struct task_struct *tsk = current;
210 	struct mm_struct *mm = tsk->mm;
211 
212 	up_read(&mm->mmap_sem);
213 	if (is_global_init(tsk)) {
214 		yield();
215 		down_read(&mm->mmap_sem);
216 		return 1;
217 	}
218 	printk("VM: killing process %s\n", tsk->comm);
219 	if (regs->psw.mask & PSW_MASK_PSTATE)
220 		do_group_exit(SIGKILL);
221 	do_no_context(regs, error_code, address);
222 	return 0;
223 }
224 
225 static void do_sigbus(struct pt_regs *regs, unsigned long error_code,
226 		      unsigned long address)
227 {
228 	struct task_struct *tsk = current;
229 	struct mm_struct *mm = tsk->mm;
230 
231 	up_read(&mm->mmap_sem);
232 	/*
233 	 * Send a sigbus, regardless of whether we were in kernel
234 	 * or user mode.
235 	 */
236 	tsk->thread.prot_addr = address;
237 	tsk->thread.trap_no = error_code;
238 	force_sig(SIGBUS, tsk);
239 
240 	/* Kernel mode? Handle exceptions or die */
241 	if (!(regs->psw.mask & PSW_MASK_PSTATE))
242 		do_no_context(regs, error_code, address);
243 }
244 
245 #ifdef CONFIG_S390_EXEC_PROTECT
246 static int signal_return(struct mm_struct *mm, struct pt_regs *regs,
247 			 unsigned long address, unsigned long error_code)
248 {
249 	u16 instruction;
250 	int rc;
251 #ifdef CONFIG_COMPAT
252 	int compat;
253 #endif
254 
255 	pagefault_disable();
256 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
257 	pagefault_enable();
258 	if (rc)
259 		return -EFAULT;
260 
261 	up_read(&mm->mmap_sem);
262 	clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
263 #ifdef CONFIG_COMPAT
264 	compat = test_tsk_thread_flag(current, TIF_31BIT);
265 	if (compat && instruction == 0x0a77)
266 		sys32_sigreturn();
267 	else if (compat && instruction == 0x0aad)
268 		sys32_rt_sigreturn();
269 	else
270 #endif
271 	if (instruction == 0x0a77)
272 		sys_sigreturn();
273 	else if (instruction == 0x0aad)
274 		sys_rt_sigreturn();
275 	else {
276 		current->thread.prot_addr = address;
277 		current->thread.trap_no = error_code;
278 		do_sigsegv(regs, error_code, SEGV_MAPERR, address);
279 	}
280 	return 0;
281 }
282 #endif /* CONFIG_S390_EXEC_PROTECT */
283 
284 /*
285  * This routine handles page faults.  It determines the address,
286  * and the problem, and then passes it off to one of the appropriate
287  * routines.
288  *
289  * error_code:
290  *   04       Protection           ->  Write-Protection  (suprression)
291  *   10       Segment translation  ->  Not present       (nullification)
292  *   11       Page translation     ->  Not present       (nullification)
293  *   3b       Region third trans.  ->  Not present       (nullification)
294  */
295 static inline void
296 do_exception(struct pt_regs *regs, unsigned long error_code, int write)
297 {
298 	struct task_struct *tsk;
299 	struct mm_struct *mm;
300 	struct vm_area_struct *vma;
301 	unsigned long address;
302 	int space;
303 	int si_code;
304 	int fault;
305 
306 	if (notify_page_fault(regs, error_code))
307 		return;
308 
309 	tsk = current;
310 	mm = tsk->mm;
311 
312 	/* get the failing address and the affected space */
313 	address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
314 	space = check_space(tsk);
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 (unlikely(space == 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 	si_code = SEGV_MAPERR;
334 	vma = find_vma(mm, address);
335 	if (!vma)
336 		goto bad_area;
337 
338 #ifdef CONFIG_S390_EXEC_PROTECT
339 	if (unlikely((space == 2) && !(vma->vm_flags & VM_EXEC)))
340 		if (!signal_return(mm, regs, address, error_code))
341 			/*
342 			 * signal_return() has done an up_read(&mm->mmap_sem)
343 			 * if it returns 0.
344 			 */
345 			return;
346 #endif
347 
348 	if (vma->vm_start <= address)
349 		goto good_area;
350 	if (!(vma->vm_flags & VM_GROWSDOWN))
351 		goto bad_area;
352 	if (expand_stack(vma, address))
353 		goto bad_area;
354 /*
355  * Ok, we have a good vm_area for this memory access, so
356  * we can handle it..
357  */
358 good_area:
359 	si_code = SEGV_ACCERR;
360 	if (!write) {
361 		/* page not present, check vm flags */
362 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
363 			goto bad_area;
364 	} else {
365 		if (!(vma->vm_flags & VM_WRITE))
366 			goto bad_area;
367 	}
368 
369 survive:
370 	/*
371 	 * If for any reason at all we couldn't handle the fault,
372 	 * make sure we exit gracefully rather than endlessly redo
373 	 * the fault.
374 	 */
375 	fault = handle_mm_fault(mm, vma, address, write);
376 	if (unlikely(fault & VM_FAULT_ERROR)) {
377 		if (fault & VM_FAULT_OOM) {
378 			if (do_out_of_memory(regs, error_code, address))
379 				goto survive;
380 			return;
381 		} else if (fault & VM_FAULT_SIGBUS) {
382 			do_sigbus(regs, error_code, address);
383 			return;
384 		}
385 		BUG();
386 	}
387 	if (fault & VM_FAULT_MAJOR)
388 		tsk->maj_flt++;
389 	else
390 		tsk->min_flt++;
391 
392         up_read(&mm->mmap_sem);
393 	/*
394 	 * The instruction that caused the program check will
395 	 * be repeated. Don't signal single step via SIGTRAP.
396 	 */
397 	clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP);
398         return;
399 
400 /*
401  * Something tried to access memory that isn't in our memory map..
402  * Fix it, but check if it's kernel or user first..
403  */
404 bad_area:
405 	up_read(&mm->mmap_sem);
406 
407 	/* User mode accesses just cause a SIGSEGV */
408 	if (regs->psw.mask & PSW_MASK_PSTATE) {
409 		tsk->thread.prot_addr = address;
410 		tsk->thread.trap_no = error_code;
411 		do_sigsegv(regs, error_code, si_code, address);
412 		return;
413 	}
414 
415 no_context:
416 	do_no_context(regs, error_code, address);
417 }
418 
419 void __kprobes do_protection_exception(struct pt_regs *regs,
420 				       long error_code)
421 {
422 	/* Protection exception is supressing, decrement psw address. */
423 	regs->psw.addr -= (error_code >> 16);
424 	/*
425 	 * Check for low-address protection.  This needs to be treated
426 	 * as a special case because the translation exception code
427 	 * field is not guaranteed to contain valid data in this case.
428 	 */
429 	if (unlikely(!(S390_lowcore.trans_exc_code & 4))) {
430 		do_low_address(regs, error_code);
431 		return;
432 	}
433 	do_exception(regs, 4, 1);
434 }
435 
436 void __kprobes do_dat_exception(struct pt_regs *regs, long error_code)
437 {
438 	do_exception(regs, error_code & 0xff, 0);
439 }
440 
441 #ifdef CONFIG_64BIT
442 void __kprobes do_asce_exception(struct pt_regs *regs, unsigned long error_code)
443 {
444 	struct mm_struct *mm;
445 	struct vm_area_struct *vma;
446 	unsigned long address;
447 	int space;
448 
449 	mm = current->mm;
450 	address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
451 	space = check_space(current);
452 
453 	if (unlikely(space == 0 || in_atomic() || !mm))
454 		goto no_context;
455 
456 	local_irq_enable();
457 
458 	down_read(&mm->mmap_sem);
459 	vma = find_vma(mm, address);
460 	up_read(&mm->mmap_sem);
461 
462 	if (vma) {
463 		update_mm(mm, current);
464 		return;
465 	}
466 
467 	/* User mode accesses just cause a SIGSEGV */
468 	if (regs->psw.mask & PSW_MASK_PSTATE) {
469 		current->thread.prot_addr = address;
470 		current->thread.trap_no = error_code;
471 		do_sigsegv(regs, error_code, SEGV_MAPERR, address);
472 		return;
473 	}
474 
475 no_context:
476 	do_no_context(regs, error_code, address);
477 }
478 #endif
479 
480 #ifdef CONFIG_PFAULT
481 /*
482  * 'pfault' pseudo page faults routines.
483  */
484 static ext_int_info_t ext_int_pfault;
485 static int pfault_disable = 0;
486 
487 static int __init nopfault(char *str)
488 {
489 	pfault_disable = 1;
490 	return 1;
491 }
492 
493 __setup("nopfault", nopfault);
494 
495 typedef struct {
496 	__u16 refdiagc;
497 	__u16 reffcode;
498 	__u16 refdwlen;
499 	__u16 refversn;
500 	__u64 refgaddr;
501 	__u64 refselmk;
502 	__u64 refcmpmk;
503 	__u64 reserved;
504 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
505 
506 int pfault_init(void)
507 {
508 	pfault_refbk_t refbk =
509 		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
510 		  __PF_RES_FIELD };
511         int rc;
512 
513 	if (!MACHINE_IS_VM || pfault_disable)
514 		return -1;
515 	asm volatile(
516 		"	diag	%1,%0,0x258\n"
517 		"0:	j	2f\n"
518 		"1:	la	%0,8\n"
519 		"2:\n"
520 		EX_TABLE(0b,1b)
521 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
522         __ctl_set_bit(0, 9);
523         return rc;
524 }
525 
526 void pfault_fini(void)
527 {
528 	pfault_refbk_t refbk =
529 	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
530 
531 	if (!MACHINE_IS_VM || pfault_disable)
532 		return;
533 	__ctl_clear_bit(0,9);
534 	asm volatile(
535 		"	diag	%0,0,0x258\n"
536 		"0:\n"
537 		EX_TABLE(0b,0b)
538 		: : "a" (&refbk), "m" (refbk) : "cc");
539 }
540 
541 static void pfault_interrupt(__u16 error_code)
542 {
543 	struct task_struct *tsk;
544 	__u16 subcode;
545 
546 	/*
547 	 * Get the external interruption subcode & pfault
548 	 * initial/completion signal bit. VM stores this
549 	 * in the 'cpu address' field associated with the
550          * external interrupt.
551 	 */
552 	subcode = S390_lowcore.cpu_addr;
553 	if ((subcode & 0xff00) != __SUBCODE_MASK)
554 		return;
555 
556 	/*
557 	 * Get the token (= address of the task structure of the affected task).
558 	 */
559 	tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
560 
561 	if (subcode & 0x0080) {
562 		/* signal bit is set -> a page has been swapped in by VM */
563 		if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
564 			/* Initial interrupt was faster than the completion
565 			 * interrupt. pfault_wait is valid. Set pfault_wait
566 			 * back to zero and wake up the process. This can
567 			 * safely be done because the task is still sleeping
568 			 * and can't produce new pfaults. */
569 			tsk->thread.pfault_wait = 0;
570 			wake_up_process(tsk);
571 			put_task_struct(tsk);
572 		}
573 	} else {
574 		/* signal bit not set -> a real page is missing. */
575 		get_task_struct(tsk);
576 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
577 		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
578 			/* Completion interrupt was faster than the initial
579 			 * interrupt (swapped in a -1 for pfault_wait). Set
580 			 * pfault_wait back to zero and exit. This can be
581 			 * done safely because tsk is running in kernel
582 			 * mode and can't produce new pfaults. */
583 			tsk->thread.pfault_wait = 0;
584 			set_task_state(tsk, TASK_RUNNING);
585 			put_task_struct(tsk);
586 		} else
587 			set_tsk_need_resched(tsk);
588 	}
589 }
590 
591 void __init pfault_irq_init(void)
592 {
593 	if (!MACHINE_IS_VM)
594 		return;
595 
596 	/*
597 	 * Try to get pfault pseudo page faults going.
598 	 */
599 	if (register_early_external_interrupt(0x2603, pfault_interrupt,
600 					      &ext_int_pfault) != 0)
601 		panic("Couldn't request external interrupt 0x2603");
602 
603 	if (pfault_init() == 0)
604 		return;
605 
606 	/* Tough luck, no pfault. */
607 	pfault_disable = 1;
608 	unregister_early_external_interrupt(0x2603, pfault_interrupt,
609 					    &ext_int_pfault);
610 }
611 #endif
612