xref: /linux/arch/s390/mm/fault.c (revision 36ca1195ad7f760a6af3814cb002bd3a3d4b4db1)
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/config.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/errno.h>
18 #include <linux/string.h>
19 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/mman.h>
22 #include <linux/mm.h>
23 #include <linux/smp.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 
30 #include <asm/system.h>
31 #include <asm/uaccess.h>
32 #include <asm/pgtable.h>
33 
34 #ifndef CONFIG_ARCH_S390X
35 #define __FAIL_ADDR_MASK 0x7ffff000
36 #define __FIXUP_MASK 0x7fffffff
37 #define __SUBCODE_MASK 0x0200
38 #define __PF_RES_FIELD 0ULL
39 #else /* CONFIG_ARCH_S390X */
40 #define __FAIL_ADDR_MASK -4096L
41 #define __FIXUP_MASK ~0L
42 #define __SUBCODE_MASK 0x0600
43 #define __PF_RES_FIELD 0x8000000000000000ULL
44 #endif /* CONFIG_ARCH_S390X */
45 
46 #ifdef CONFIG_SYSCTL
47 extern int sysctl_userprocess_debug;
48 #endif
49 
50 extern void die(const char *,struct pt_regs *,long);
51 
52 extern spinlock_t timerlist_lock;
53 
54 /*
55  * Unlock any spinlocks which will prevent us from getting the
56  * message out (timerlist_lock is acquired through the
57  * console unblank code)
58  */
59 void bust_spinlocks(int yes)
60 {
61 	if (yes) {
62 		oops_in_progress = 1;
63 	} else {
64 		int loglevel_save = console_loglevel;
65 		console_unblank();
66 		oops_in_progress = 0;
67 		/*
68 		 * OK, the message is on the console.  Now we call printk()
69 		 * without oops_in_progress set so that printk will give klogd
70 		 * a poke.  Hold onto your hats...
71 		 */
72 		console_loglevel = 15;
73 		printk(" ");
74 		console_loglevel = loglevel_save;
75 	}
76 }
77 
78 /*
79  * Check which address space is addressed by the access
80  * register in S390_lowcore.exc_access_id.
81  * Returns 1 for user space and 0 for kernel space.
82  */
83 static int __check_access_register(struct pt_regs *regs, int error_code)
84 {
85 	int areg = S390_lowcore.exc_access_id;
86 
87 	if (areg == 0)
88 		/* Access via access register 0 -> kernel address */
89 		return 0;
90 	save_access_regs(current->thread.acrs);
91 	if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
92 		/*
93 		 * access register contains 0 -> kernel address,
94 		 * access register contains 1 -> user space address
95 		 */
96 		return current->thread.acrs[areg];
97 
98 	/* Something unhealthy was done with the access registers... */
99 	die("page fault via unknown access register", regs, error_code);
100 	do_exit(SIGKILL);
101 	return 0;
102 }
103 
104 /*
105  * Check which address space the address belongs to.
106  * Returns 1 for user space and 0 for kernel space.
107  */
108 static inline int check_user_space(struct pt_regs *regs, int error_code)
109 {
110 	/*
111 	 * The lowest two bits of S390_lowcore.trans_exc_code indicate
112 	 * which paging table was used:
113 	 *   0: Primary Segment Table Descriptor
114 	 *   1: STD determined via access register
115 	 *   2: Secondary Segment Table Descriptor
116 	 *   3: Home Segment Table Descriptor
117 	 */
118 	int descriptor = S390_lowcore.trans_exc_code & 3;
119 	if (unlikely(descriptor == 1))
120 		return __check_access_register(regs, error_code);
121 	if (descriptor == 2)
122 		return current->thread.mm_segment.ar4;
123 	return descriptor != 0;
124 }
125 
126 /*
127  * Send SIGSEGV to task.  This is an external routine
128  * to keep the stack usage of do_page_fault small.
129  */
130 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
131 		       int si_code, unsigned long address)
132 {
133 	struct siginfo si;
134 
135 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
136 #if defined(CONFIG_SYSCTL)
137 	if (sysctl_userprocess_debug)
138 #endif
139 	{
140 		printk("User process fault: interruption code 0x%lX\n",
141 		       error_code);
142 		printk("failing address: %lX\n", address);
143 		show_regs(regs);
144 	}
145 #endif
146 	si.si_signo = SIGSEGV;
147 	si.si_code = si_code;
148 	si.si_addr = (void *) address;
149 	force_sig_info(SIGSEGV, &si, current);
150 }
151 
152 /*
153  * This routine handles page faults.  It determines the address,
154  * and the problem, and then passes it off to one of the appropriate
155  * routines.
156  *
157  * error_code:
158  *   04       Protection           ->  Write-Protection  (suprression)
159  *   10       Segment translation  ->  Not present       (nullification)
160  *   11       Page translation     ->  Not present       (nullification)
161  *   3b       Region third trans.  ->  Not present       (nullification)
162  */
163 extern inline void
164 do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection)
165 {
166         struct task_struct *tsk;
167         struct mm_struct *mm;
168         struct vm_area_struct * vma;
169         unsigned long address;
170 	int user_address;
171 	const struct exception_table_entry *fixup;
172 	int si_code = SEGV_MAPERR;
173 
174         tsk = current;
175         mm = tsk->mm;
176 
177 	/*
178          * Check for low-address protection.  This needs to be treated
179 	 * as a special case because the translation exception code
180 	 * field is not guaranteed to contain valid data in this case.
181 	 */
182 	if (is_protection && !(S390_lowcore.trans_exc_code & 4)) {
183 
184 		/* Low-address protection hit in kernel mode means
185 		   NULL pointer write access in kernel mode.  */
186  		if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
187 			address = 0;
188 			user_address = 0;
189 			goto no_context;
190 		}
191 
192 		/* Low-address protection hit in user mode 'cannot happen'.  */
193 		die ("Low-address protection", regs, error_code);
194         	do_exit(SIGKILL);
195 	}
196 
197         /*
198          * get the failing address
199          * more specific the segment and page table portion of
200          * the address
201          */
202         address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
203 	user_address = check_user_space(regs, error_code);
204 
205 	/*
206 	 * Verify that the fault happened in user space, that
207 	 * we are not in an interrupt and that there is a
208 	 * user context.
209 	 */
210         if (user_address == 0 || in_atomic() || !mm)
211                 goto no_context;
212 
213 	/*
214 	 * When we get here, the fault happened in the current
215 	 * task's user address space, so we can switch on the
216 	 * interrupts again and then search the VMAs
217 	 */
218 	local_irq_enable();
219 
220         down_read(&mm->mmap_sem);
221 
222         vma = find_vma(mm, address);
223         if (!vma)
224                 goto bad_area;
225         if (vma->vm_start <= address)
226                 goto good_area;
227         if (!(vma->vm_flags & VM_GROWSDOWN))
228                 goto bad_area;
229         if (expand_stack(vma, address))
230                 goto bad_area;
231 /*
232  * Ok, we have a good vm_area for this memory access, so
233  * we can handle it..
234  */
235 good_area:
236 	si_code = SEGV_ACCERR;
237 	if (!is_protection) {
238 		/* page not present, check vm flags */
239 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
240 			goto bad_area;
241 	} else {
242 		if (!(vma->vm_flags & VM_WRITE))
243 			goto bad_area;
244 	}
245 
246 survive:
247 	/*
248 	 * If for any reason at all we couldn't handle the fault,
249 	 * make sure we exit gracefully rather than endlessly redo
250 	 * the fault.
251 	 */
252 	switch (handle_mm_fault(mm, vma, address, is_protection)) {
253 	case VM_FAULT_MINOR:
254 		tsk->min_flt++;
255 		break;
256 	case VM_FAULT_MAJOR:
257 		tsk->maj_flt++;
258 		break;
259 	case VM_FAULT_SIGBUS:
260 		goto do_sigbus;
261 	case VM_FAULT_OOM:
262 		goto out_of_memory;
263 	default:
264 		BUG();
265 	}
266 
267         up_read(&mm->mmap_sem);
268 	/*
269 	 * The instruction that caused the program check will
270 	 * be repeated. Don't signal single step via SIGTRAP.
271 	 */
272 	clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
273         return;
274 
275 /*
276  * Something tried to access memory that isn't in our memory map..
277  * Fix it, but check if it's kernel or user first..
278  */
279 bad_area:
280         up_read(&mm->mmap_sem);
281 
282         /* User mode accesses just cause a SIGSEGV */
283         if (regs->psw.mask & PSW_MASK_PSTATE) {
284                 tsk->thread.prot_addr = address;
285                 tsk->thread.trap_no = error_code;
286 		do_sigsegv(regs, error_code, si_code, address);
287                 return;
288 	}
289 
290 no_context:
291         /* Are we prepared to handle this kernel fault?  */
292 	fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
293 	if (fixup) {
294 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
295                 return;
296         }
297 
298 /*
299  * Oops. The kernel tried to access some bad page. We'll have to
300  * terminate things with extreme prejudice.
301  */
302         if (user_address == 0)
303                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
304         	       " at virtual kernel address %p\n", (void *)address);
305         else
306                 printk(KERN_ALERT "Unable to handle kernel paging request"
307 		       " at virtual user address %p\n", (void *)address);
308 
309         die("Oops", regs, error_code);
310         do_exit(SIGKILL);
311 
312 
313 /*
314  * We ran out of memory, or some other thing happened to us that made
315  * us unable to handle the page fault gracefully.
316 */
317 out_of_memory:
318 	up_read(&mm->mmap_sem);
319 	if (tsk->pid == 1) {
320 		yield();
321 		goto survive;
322 	}
323 	printk("VM: killing process %s\n", tsk->comm);
324 	if (regs->psw.mask & PSW_MASK_PSTATE)
325 		do_exit(SIGKILL);
326 	goto no_context;
327 
328 do_sigbus:
329 	up_read(&mm->mmap_sem);
330 
331 	/*
332 	 * Send a sigbus, regardless of whether we were in kernel
333 	 * or user mode.
334 	 */
335         tsk->thread.prot_addr = address;
336         tsk->thread.trap_no = error_code;
337 	force_sig(SIGBUS, tsk);
338 
339 	/* Kernel mode? Handle exceptions or die */
340 	if (!(regs->psw.mask & PSW_MASK_PSTATE))
341 		goto no_context;
342 }
343 
344 void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
345 {
346 	regs->psw.addr -= (error_code >> 16);
347 	do_exception(regs, 4, 1);
348 }
349 
350 void do_dat_exception(struct pt_regs *regs, unsigned long error_code)
351 {
352 	do_exception(regs, error_code & 0xff, 0);
353 }
354 
355 #ifndef CONFIG_ARCH_S390X
356 
357 typedef struct _pseudo_wait_t {
358        struct _pseudo_wait_t *next;
359        wait_queue_head_t queue;
360        unsigned long address;
361        int resolved;
362 } pseudo_wait_t;
363 
364 static pseudo_wait_t *pseudo_lock_queue = NULL;
365 static spinlock_t pseudo_wait_spinlock; /* spinlock to protect lock queue */
366 
367 /*
368  * This routine handles 'pagex' pseudo page faults.
369  */
370 asmlinkage void
371 do_pseudo_page_fault(struct pt_regs *regs, unsigned long error_code)
372 {
373         pseudo_wait_t wait_struct;
374         pseudo_wait_t *ptr, *last, *next;
375         unsigned long address;
376 
377         /*
378          * get the failing address
379          * more specific the segment and page table portion of
380          * the address
381          */
382         address = S390_lowcore.trans_exc_code & 0xfffff000;
383 
384         if (address & 0x80000000) {
385                 /* high bit set -> a page has been swapped in by VM */
386                 address &= 0x7fffffff;
387                 spin_lock(&pseudo_wait_spinlock);
388                 last = NULL;
389                 ptr = pseudo_lock_queue;
390                 while (ptr != NULL) {
391                         next = ptr->next;
392                         if (address == ptr->address) {
393 				 /*
394                                  * This is one of the processes waiting
395                                  * for the page. Unchain from the queue.
396                                  * There can be more than one process
397                                  * waiting for the same page. VM presents
398                                  * an initial and a completion interrupt for
399                                  * every process that tries to access a
400                                  * page swapped out by VM.
401                                  */
402                                 if (last == NULL)
403                                         pseudo_lock_queue = next;
404                                 else
405                                         last->next = next;
406                                 /* now wake up the process */
407                                 ptr->resolved = 1;
408                                 wake_up(&ptr->queue);
409                         } else
410                                 last = ptr;
411                         ptr = next;
412                 }
413                 spin_unlock(&pseudo_wait_spinlock);
414         } else {
415                 /* Pseudo page faults in kernel mode is a bad idea */
416                 if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
417                         /*
418 			 * VM presents pseudo page faults if the interrupted
419 			 * state was not disabled for interrupts. So we can
420 			 * get pseudo page fault interrupts while running
421 			 * in kernel mode. We simply access the page here
422 			 * while we are running disabled. VM will then swap
423 			 * in the page synchronously.
424                          */
425                          if (check_user_space(regs, error_code) == 0)
426                                  /* dereference a virtual kernel address */
427                                  __asm__ __volatile__ (
428                                          "  ic 0,0(%0)"
429                                          : : "a" (address) : "0");
430                          else
431                                  /* dereference a virtual user address */
432                                  __asm__ __volatile__ (
433                                          "  la   2,0(%0)\n"
434                                          "  sacf 512\n"
435                                          "  ic   2,0(2)\n"
436 					 "0:sacf 0\n"
437 					 ".section __ex_table,\"a\"\n"
438 					 "  .align 4\n"
439 					 "  .long  0b,0b\n"
440 					 ".previous"
441                                          : : "a" (address) : "2" );
442 
443                         return;
444                 }
445 		/* initialize and add element to pseudo_lock_queue */
446                 init_waitqueue_head (&wait_struct.queue);
447                 wait_struct.address = address;
448                 wait_struct.resolved = 0;
449                 spin_lock(&pseudo_wait_spinlock);
450                 wait_struct.next = pseudo_lock_queue;
451                 pseudo_lock_queue = &wait_struct;
452                 spin_unlock(&pseudo_wait_spinlock);
453 		/*
454 		 * The instruction that caused the program check will
455 		 * be repeated. Don't signal single step via SIGTRAP.
456 		 */
457 		clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
458                 /* go to sleep */
459                 wait_event(wait_struct.queue, wait_struct.resolved);
460         }
461 }
462 #endif /* CONFIG_ARCH_S390X */
463 
464 #ifdef CONFIG_PFAULT
465 /*
466  * 'pfault' pseudo page faults routines.
467  */
468 static int pfault_disable = 0;
469 
470 static int __init nopfault(char *str)
471 {
472 	pfault_disable = 1;
473 	return 1;
474 }
475 
476 __setup("nopfault", nopfault);
477 
478 typedef struct {
479 	__u16 refdiagc;
480 	__u16 reffcode;
481 	__u16 refdwlen;
482 	__u16 refversn;
483 	__u64 refgaddr;
484 	__u64 refselmk;
485 	__u64 refcmpmk;
486 	__u64 reserved;
487 } __attribute__ ((packed)) pfault_refbk_t;
488 
489 int pfault_init(void)
490 {
491 	pfault_refbk_t refbk =
492 		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
493 		  __PF_RES_FIELD };
494         int rc;
495 
496 	if (pfault_disable)
497 		return -1;
498         __asm__ __volatile__(
499                 "    diag  %1,%0,0x258\n"
500 		"0:  j     2f\n"
501 		"1:  la    %0,8\n"
502 		"2:\n"
503 		".section __ex_table,\"a\"\n"
504 		"   .align 4\n"
505 #ifndef CONFIG_ARCH_S390X
506 		"   .long  0b,1b\n"
507 #else /* CONFIG_ARCH_S390X */
508 		"   .quad  0b,1b\n"
509 #endif /* CONFIG_ARCH_S390X */
510 		".previous"
511                 : "=d" (rc) : "a" (&refbk) : "cc" );
512         __ctl_set_bit(0, 9);
513         return rc;
514 }
515 
516 void pfault_fini(void)
517 {
518 	pfault_refbk_t refbk =
519 	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
520 
521 	if (pfault_disable)
522 		return;
523 	__ctl_clear_bit(0,9);
524         __asm__ __volatile__(
525                 "    diag  %0,0,0x258\n"
526 		"0:\n"
527 		".section __ex_table,\"a\"\n"
528 		"   .align 4\n"
529 #ifndef CONFIG_ARCH_S390X
530 		"   .long  0b,0b\n"
531 #else /* CONFIG_ARCH_S390X */
532 		"   .quad  0b,0b\n"
533 #endif /* CONFIG_ARCH_S390X */
534 		".previous"
535 		: : "a" (&refbk) : "cc" );
536 }
537 
538 asmlinkage void
539 pfault_interrupt(struct pt_regs *regs, __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 procude new pfaults. */
567 			tsk->thread.pfault_wait = 0;
568 			wake_up_process(tsk);
569 		}
570 	} else {
571 		/* signal bit not set -> a real page is missing. */
572 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
573 		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
574 			/* Completion interrupt was faster than the initial
575 			 * interrupt (swapped in a -1 for pfault_wait). Set
576 			 * pfault_wait back to zero and exit. This can be
577 			 * done safely because tsk is running in kernel
578 			 * mode and can't produce new pfaults. */
579 			tsk->thread.pfault_wait = 0;
580 			set_task_state(tsk, TASK_RUNNING);
581 		} else
582 			set_tsk_need_resched(tsk);
583 	}
584 }
585 #endif
586 
587