xref: /linux/arch/s390/mm/fault.c (revision 6e8331ac6973435b1e7604c30f2ad394035b46e1)
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 
29 #include <asm/system.h>
30 #include <asm/uaccess.h>
31 #include <asm/pgtable.h>
32 
33 #ifndef CONFIG_64BIT
34 #define __FAIL_ADDR_MASK 0x7ffff000
35 #define __FIXUP_MASK 0x7fffffff
36 #define __SUBCODE_MASK 0x0200
37 #define __PF_RES_FIELD 0ULL
38 #else /* CONFIG_64BIT */
39 #define __FAIL_ADDR_MASK -4096L
40 #define __FIXUP_MASK ~0L
41 #define __SUBCODE_MASK 0x0600
42 #define __PF_RES_FIELD 0x8000000000000000ULL
43 #endif /* CONFIG_64BIT */
44 
45 #ifdef CONFIG_SYSCTL
46 extern int sysctl_userprocess_debug;
47 #endif
48 
49 extern void die(const char *,struct pt_regs *,long);
50 
51 extern spinlock_t timerlist_lock;
52 
53 /*
54  * Unlock any spinlocks which will prevent us from getting the
55  * message out (timerlist_lock is acquired through the
56  * console unblank code)
57  */
58 void bust_spinlocks(int yes)
59 {
60 	if (yes) {
61 		oops_in_progress = 1;
62 	} else {
63 		int loglevel_save = console_loglevel;
64 		console_unblank();
65 		oops_in_progress = 0;
66 		/*
67 		 * OK, the message is on the console.  Now we call printk()
68 		 * without oops_in_progress set so that printk will give klogd
69 		 * a poke.  Hold onto your hats...
70 		 */
71 		console_loglevel = 15;
72 		printk(" ");
73 		console_loglevel = loglevel_save;
74 	}
75 }
76 
77 /*
78  * Check which address space is addressed by the access
79  * register in S390_lowcore.exc_access_id.
80  * Returns 1 for user space and 0 for kernel space.
81  */
82 static int __check_access_register(struct pt_regs *regs, int error_code)
83 {
84 	int areg = S390_lowcore.exc_access_id;
85 
86 	if (areg == 0)
87 		/* Access via access register 0 -> kernel address */
88 		return 0;
89 	save_access_regs(current->thread.acrs);
90 	if (regs && areg < NUM_ACRS && current->thread.acrs[areg] <= 1)
91 		/*
92 		 * access register contains 0 -> kernel address,
93 		 * access register contains 1 -> user space address
94 		 */
95 		return current->thread.acrs[areg];
96 
97 	/* Something unhealthy was done with the access registers... */
98 	die("page fault via unknown access register", regs, error_code);
99 	do_exit(SIGKILL);
100 	return 0;
101 }
102 
103 /*
104  * Check which address space the address belongs to.
105  * Returns 1 for user space and 0 for kernel space.
106  */
107 static inline int check_user_space(struct pt_regs *regs, int error_code)
108 {
109 	/*
110 	 * The lowest two bits of S390_lowcore.trans_exc_code indicate
111 	 * which paging table was used:
112 	 *   0: Primary Segment Table Descriptor
113 	 *   1: STD determined via access register
114 	 *   2: Secondary Segment Table Descriptor
115 	 *   3: Home Segment Table Descriptor
116 	 */
117 	int descriptor = S390_lowcore.trans_exc_code & 3;
118 	if (unlikely(descriptor == 1))
119 		return __check_access_register(regs, error_code);
120 	if (descriptor == 2)
121 		return current->thread.mm_segment.ar4;
122 	return descriptor != 0;
123 }
124 
125 /*
126  * Send SIGSEGV to task.  This is an external routine
127  * to keep the stack usage of do_page_fault small.
128  */
129 static void do_sigsegv(struct pt_regs *regs, unsigned long error_code,
130 		       int si_code, unsigned long address)
131 {
132 	struct siginfo si;
133 
134 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
135 #if defined(CONFIG_SYSCTL)
136 	if (sysctl_userprocess_debug)
137 #endif
138 	{
139 		printk("User process fault: interruption code 0x%lX\n",
140 		       error_code);
141 		printk("failing address: %lX\n", address);
142 		show_regs(regs);
143 	}
144 #endif
145 	si.si_signo = SIGSEGV;
146 	si.si_code = si_code;
147 	si.si_addr = (void __user *) address;
148 	force_sig_info(SIGSEGV, &si, current);
149 }
150 
151 /*
152  * This routine handles page faults.  It determines the address,
153  * and the problem, and then passes it off to one of the appropriate
154  * routines.
155  *
156  * error_code:
157  *   04       Protection           ->  Write-Protection  (suprression)
158  *   10       Segment translation  ->  Not present       (nullification)
159  *   11       Page translation     ->  Not present       (nullification)
160  *   3b       Region third trans.  ->  Not present       (nullification)
161  */
162 static inline void
163 do_exception(struct pt_regs *regs, unsigned long error_code, int is_protection)
164 {
165         struct task_struct *tsk;
166         struct mm_struct *mm;
167         struct vm_area_struct * vma;
168         unsigned long address;
169 	int user_address;
170 	const struct exception_table_entry *fixup;
171 	int si_code = SEGV_MAPERR;
172 
173         tsk = current;
174         mm = tsk->mm;
175 
176 	/*
177          * Check for low-address protection.  This needs to be treated
178 	 * as a special case because the translation exception code
179 	 * field is not guaranteed to contain valid data in this case.
180 	 */
181 	if (is_protection && !(S390_lowcore.trans_exc_code & 4)) {
182 
183 		/* Low-address protection hit in kernel mode means
184 		   NULL pointer write access in kernel mode.  */
185  		if (!(regs->psw.mask & PSW_MASK_PSTATE)) {
186 			address = 0;
187 			user_address = 0;
188 			goto no_context;
189 		}
190 
191 		/* Low-address protection hit in user mode 'cannot happen'.  */
192 		die ("Low-address protection", regs, error_code);
193         	do_exit(SIGKILL);
194 	}
195 
196         /*
197          * get the failing address
198          * more specific the segment and page table portion of
199          * the address
200          */
201         address = S390_lowcore.trans_exc_code & __FAIL_ADDR_MASK;
202 	user_address = check_user_space(regs, error_code);
203 
204 	/*
205 	 * Verify that the fault happened in user space, that
206 	 * we are not in an interrupt and that there is a
207 	 * user context.
208 	 */
209         if (user_address == 0 || in_atomic() || !mm)
210                 goto no_context;
211 
212 	/*
213 	 * When we get here, the fault happened in the current
214 	 * task's user address space, so we can switch on the
215 	 * interrupts again and then search the VMAs
216 	 */
217 	local_irq_enable();
218 
219         down_read(&mm->mmap_sem);
220 
221         vma = find_vma(mm, address);
222         if (!vma)
223                 goto bad_area;
224         if (vma->vm_start <= address)
225                 goto good_area;
226         if (!(vma->vm_flags & VM_GROWSDOWN))
227                 goto bad_area;
228         if (expand_stack(vma, address))
229                 goto bad_area;
230 /*
231  * Ok, we have a good vm_area for this memory access, so
232  * we can handle it..
233  */
234 good_area:
235 	si_code = SEGV_ACCERR;
236 	if (!is_protection) {
237 		/* page not present, check vm flags */
238 		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
239 			goto bad_area;
240 	} else {
241 		if (!(vma->vm_flags & VM_WRITE))
242 			goto bad_area;
243 	}
244 
245 survive:
246 	/*
247 	 * If for any reason at all we couldn't handle the fault,
248 	 * make sure we exit gracefully rather than endlessly redo
249 	 * the fault.
250 	 */
251 	switch (handle_mm_fault(mm, vma, address, is_protection)) {
252 	case VM_FAULT_MINOR:
253 		tsk->min_flt++;
254 		break;
255 	case VM_FAULT_MAJOR:
256 		tsk->maj_flt++;
257 		break;
258 	case VM_FAULT_SIGBUS:
259 		goto do_sigbus;
260 	case VM_FAULT_OOM:
261 		goto out_of_memory;
262 	default:
263 		BUG();
264 	}
265 
266         up_read(&mm->mmap_sem);
267 	/*
268 	 * The instruction that caused the program check will
269 	 * be repeated. Don't signal single step via SIGTRAP.
270 	 */
271 	clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
272         return;
273 
274 /*
275  * Something tried to access memory that isn't in our memory map..
276  * Fix it, but check if it's kernel or user first..
277  */
278 bad_area:
279         up_read(&mm->mmap_sem);
280 
281         /* User mode accesses just cause a SIGSEGV */
282         if (regs->psw.mask & PSW_MASK_PSTATE) {
283                 tsk->thread.prot_addr = address;
284                 tsk->thread.trap_no = error_code;
285 		do_sigsegv(regs, error_code, si_code, address);
286                 return;
287 	}
288 
289 no_context:
290         /* Are we prepared to handle this kernel fault?  */
291 	fixup = search_exception_tables(regs->psw.addr & __FIXUP_MASK);
292 	if (fixup) {
293 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
294                 return;
295         }
296 
297 /*
298  * Oops. The kernel tried to access some bad page. We'll have to
299  * terminate things with extreme prejudice.
300  */
301         if (user_address == 0)
302                 printk(KERN_ALERT "Unable to handle kernel pointer dereference"
303         	       " at virtual kernel address %p\n", (void *)address);
304         else
305                 printk(KERN_ALERT "Unable to handle kernel paging request"
306 		       " at virtual user address %p\n", (void *)address);
307 
308         die("Oops", regs, error_code);
309         do_exit(SIGKILL);
310 
311 
312 /*
313  * We ran out of memory, or some other thing happened to us that made
314  * us unable to handle the page fault gracefully.
315 */
316 out_of_memory:
317 	up_read(&mm->mmap_sem);
318 	if (tsk->pid == 1) {
319 		yield();
320 		goto survive;
321 	}
322 	printk("VM: killing process %s\n", tsk->comm);
323 	if (regs->psw.mask & PSW_MASK_PSTATE)
324 		do_exit(SIGKILL);
325 	goto no_context;
326 
327 do_sigbus:
328 	up_read(&mm->mmap_sem);
329 
330 	/*
331 	 * Send a sigbus, regardless of whether we were in kernel
332 	 * or user mode.
333 	 */
334         tsk->thread.prot_addr = address;
335         tsk->thread.trap_no = error_code;
336 	force_sig(SIGBUS, tsk);
337 
338 	/* Kernel mode? Handle exceptions or die */
339 	if (!(regs->psw.mask & PSW_MASK_PSTATE))
340 		goto no_context;
341 }
342 
343 void do_protection_exception(struct pt_regs *regs, unsigned long error_code)
344 {
345 	regs->psw.addr -= (error_code >> 16);
346 	do_exception(regs, 4, 1);
347 }
348 
349 void do_dat_exception(struct pt_regs *regs, unsigned long error_code)
350 {
351 	do_exception(regs, error_code & 0xff, 0);
352 }
353 
354 #ifdef CONFIG_PFAULT
355 /*
356  * 'pfault' pseudo page faults routines.
357  */
358 static int pfault_disable = 0;
359 
360 static int __init nopfault(char *str)
361 {
362 	pfault_disable = 1;
363 	return 1;
364 }
365 
366 __setup("nopfault", nopfault);
367 
368 typedef struct {
369 	__u16 refdiagc;
370 	__u16 reffcode;
371 	__u16 refdwlen;
372 	__u16 refversn;
373 	__u64 refgaddr;
374 	__u64 refselmk;
375 	__u64 refcmpmk;
376 	__u64 reserved;
377 } __attribute__ ((packed)) pfault_refbk_t;
378 
379 int pfault_init(void)
380 {
381 	pfault_refbk_t refbk =
382 		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
383 		  __PF_RES_FIELD };
384         int rc;
385 
386 	if (pfault_disable)
387 		return -1;
388         __asm__ __volatile__(
389                 "    diag  %1,%0,0x258\n"
390 		"0:  j     2f\n"
391 		"1:  la    %0,8\n"
392 		"2:\n"
393 		".section __ex_table,\"a\"\n"
394 		"   .align 4\n"
395 #ifndef CONFIG_64BIT
396 		"   .long  0b,1b\n"
397 #else /* CONFIG_64BIT */
398 		"   .quad  0b,1b\n"
399 #endif /* CONFIG_64BIT */
400 		".previous"
401                 : "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc" );
402         __ctl_set_bit(0, 9);
403         return rc;
404 }
405 
406 void pfault_fini(void)
407 {
408 	pfault_refbk_t refbk =
409 	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
410 
411 	if (pfault_disable)
412 		return;
413 	__ctl_clear_bit(0,9);
414         __asm__ __volatile__(
415                 "    diag  %0,0,0x258\n"
416 		"0:\n"
417 		".section __ex_table,\"a\"\n"
418 		"   .align 4\n"
419 #ifndef CONFIG_64BIT
420 		"   .long  0b,0b\n"
421 #else /* CONFIG_64BIT */
422 		"   .quad  0b,0b\n"
423 #endif /* CONFIG_64BIT */
424 		".previous"
425 		: : "a" (&refbk), "m" (refbk) : "cc" );
426 }
427 
428 asmlinkage void
429 pfault_interrupt(struct pt_regs *regs, __u16 error_code)
430 {
431 	struct task_struct *tsk;
432 	__u16 subcode;
433 
434 	/*
435 	 * Get the external interruption subcode & pfault
436 	 * initial/completion signal bit. VM stores this
437 	 * in the 'cpu address' field associated with the
438          * external interrupt.
439 	 */
440 	subcode = S390_lowcore.cpu_addr;
441 	if ((subcode & 0xff00) != __SUBCODE_MASK)
442 		return;
443 
444 	/*
445 	 * Get the token (= address of the task structure of the affected task).
446 	 */
447 	tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
448 
449 	if (subcode & 0x0080) {
450 		/* signal bit is set -> a page has been swapped in by VM */
451 		if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
452 			/* Initial interrupt was faster than the completion
453 			 * interrupt. pfault_wait is valid. Set pfault_wait
454 			 * back to zero and wake up the process. This can
455 			 * safely be done because the task is still sleeping
456 			 * and can't produce new pfaults. */
457 			tsk->thread.pfault_wait = 0;
458 			wake_up_process(tsk);
459 			put_task_struct(tsk);
460 		}
461 	} else {
462 		/* signal bit not set -> a real page is missing. */
463 		get_task_struct(tsk);
464 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
465 		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
466 			/* Completion interrupt was faster than the initial
467 			 * interrupt (swapped in a -1 for pfault_wait). Set
468 			 * pfault_wait back to zero and exit. This can be
469 			 * done safely because tsk is running in kernel
470 			 * mode and can't produce new pfaults. */
471 			tsk->thread.pfault_wait = 0;
472 			set_task_state(tsk, TASK_RUNNING);
473 			put_task_struct(tsk);
474 		} else
475 			set_tsk_need_resched(tsk);
476 	}
477 }
478 #endif
479 
480