xref: /openbmc/linux/arch/s390/mm/fault.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
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/perf_event.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/compat.h>
24 #include <linux/smp.h>
25 #include <linux/kdebug.h>
26 #include <linux/init.h>
27 #include <linux/console.h>
28 #include <linux/module.h>
29 #include <linux/hardirq.h>
30 #include <linux/kprobes.h>
31 #include <linux/uaccess.h>
32 #include <linux/hugetlb.h>
33 #include <asm/system.h>
34 #include <asm/pgtable.h>
35 #include <asm/s390_ext.h>
36 #include <asm/mmu_context.h>
37 #include <asm/compat.h>
38 #include "../kernel/entry.h"
39 
40 #ifndef CONFIG_64BIT
41 #define __FAIL_ADDR_MASK 0x7ffff000
42 #define __SUBCODE_MASK 0x0200
43 #define __PF_RES_FIELD 0ULL
44 #else /* CONFIG_64BIT */
45 #define __FAIL_ADDR_MASK -4096L
46 #define __SUBCODE_MASK 0x0600
47 #define __PF_RES_FIELD 0x8000000000000000ULL
48 #endif /* CONFIG_64BIT */
49 
50 #ifdef CONFIG_SYSCTL
51 extern int sysctl_userprocess_debug;
52 #endif
53 
54 #define VM_FAULT_BADCONTEXT	0x010000
55 #define VM_FAULT_BADMAP		0x020000
56 #define VM_FAULT_BADACCESS	0x040000
57 
58 static inline int notify_page_fault(struct pt_regs *regs)
59 {
60 	int ret = 0;
61 
62 #ifdef CONFIG_KPROBES
63 	/* kprobe_running() needs smp_processor_id() */
64 	if (!user_mode(regs)) {
65 		preempt_disable();
66 		if (kprobe_running() && kprobe_fault_handler(regs, 14))
67 			ret = 1;
68 		preempt_enable();
69 	}
70 #endif
71 	return ret;
72 }
73 
74 
75 /*
76  * Unlock any spinlocks which will prevent us from getting the
77  * message out.
78  */
79 void bust_spinlocks(int yes)
80 {
81 	if (yes) {
82 		oops_in_progress = 1;
83 	} else {
84 		int loglevel_save = console_loglevel;
85 		console_unblank();
86 		oops_in_progress = 0;
87 		/*
88 		 * OK, the message is on the console.  Now we call printk()
89 		 * without oops_in_progress set so that printk will give klogd
90 		 * a poke.  Hold onto your hats...
91 		 */
92 		console_loglevel = 15;
93 		printk(" ");
94 		console_loglevel = loglevel_save;
95 	}
96 }
97 
98 /*
99  * Returns the address space associated with the fault.
100  * Returns 0 for kernel space and 1 for user space.
101  */
102 static inline int user_space_fault(unsigned long trans_exc_code)
103 {
104 	/*
105 	 * The lowest two bits of the translation exception
106 	 * identification indicate which paging table was used.
107 	 */
108 	trans_exc_code &= 3;
109 	if (trans_exc_code == 2)
110 		/* Access via secondary space, set_fs setting decides */
111 		return current->thread.mm_segment.ar4;
112 	if (user_mode == HOME_SPACE_MODE)
113 		/* User space if the access has been done via home space. */
114 		return trans_exc_code == 3;
115 	/*
116 	 * If the user space is not the home space the kernel runs in home
117 	 * space. Access via secondary space has already been covered,
118 	 * access via primary space or access register is from user space
119 	 * and access via home space is from the kernel.
120 	 */
121 	return trans_exc_code != 3;
122 }
123 
124 /*
125  * Send SIGSEGV to task.  This is an external routine
126  * to keep the stack usage of do_page_fault small.
127  */
128 static noinline void do_sigsegv(struct pt_regs *regs, long int_code,
129 				int si_code, unsigned long trans_exc_code)
130 {
131 	struct siginfo si;
132 	unsigned long address;
133 
134 	address = trans_exc_code & __FAIL_ADDR_MASK;
135 	current->thread.prot_addr = address;
136 	current->thread.trap_no = int_code;
137 #if defined(CONFIG_SYSCTL) || defined(CONFIG_PROCESS_DEBUG)
138 #if defined(CONFIG_SYSCTL)
139 	if (sysctl_userprocess_debug)
140 #endif
141 	{
142 		printk("User process fault: interruption code 0x%lX\n",
143 		       int_code);
144 		printk("failing address: %lX\n", address);
145 		show_regs(regs);
146 	}
147 #endif
148 	si.si_signo = SIGSEGV;
149 	si.si_code = si_code;
150 	si.si_addr = (void __user *) address;
151 	force_sig_info(SIGSEGV, &si, current);
152 }
153 
154 static noinline void do_no_context(struct pt_regs *regs, long int_code,
155 				   unsigned long trans_exc_code)
156 {
157 	const struct exception_table_entry *fixup;
158 	unsigned long address;
159 
160 	/* Are we prepared to handle this kernel fault?  */
161 	fixup = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
162 	if (fixup) {
163 		regs->psw.addr = fixup->fixup | PSW_ADDR_AMODE;
164 		return;
165 	}
166 
167 	/*
168 	 * Oops. The kernel tried to access some bad page. We'll have to
169 	 * terminate things with extreme prejudice.
170 	 */
171 	address = trans_exc_code & __FAIL_ADDR_MASK;
172 	if (!user_space_fault(trans_exc_code))
173 		printk(KERN_ALERT "Unable to handle kernel pointer dereference"
174 		       " at virtual kernel address %p\n", (void *)address);
175 	else
176 		printk(KERN_ALERT "Unable to handle kernel paging request"
177 		       " at virtual user address %p\n", (void *)address);
178 
179 	die("Oops", regs, int_code);
180 	do_exit(SIGKILL);
181 }
182 
183 static noinline void do_low_address(struct pt_regs *regs, long int_code,
184 				    unsigned long trans_exc_code)
185 {
186 	/* Low-address protection hit in kernel mode means
187 	   NULL pointer write access in kernel mode.  */
188 	if (regs->psw.mask & PSW_MASK_PSTATE) {
189 		/* Low-address protection hit in user mode 'cannot happen'. */
190 		die ("Low-address protection", regs, int_code);
191 		do_exit(SIGKILL);
192 	}
193 
194 	do_no_context(regs, int_code, trans_exc_code);
195 }
196 
197 static noinline void do_sigbus(struct pt_regs *regs, long int_code,
198 			       unsigned long trans_exc_code)
199 {
200 	struct task_struct *tsk = current;
201 
202 	/*
203 	 * Send a sigbus, regardless of whether we were in kernel
204 	 * or user mode.
205 	 */
206 	tsk->thread.prot_addr = trans_exc_code & __FAIL_ADDR_MASK;
207 	tsk->thread.trap_no = int_code;
208 	force_sig(SIGBUS, tsk);
209 }
210 
211 #ifdef CONFIG_S390_EXEC_PROTECT
212 static noinline int signal_return(struct pt_regs *regs, long int_code,
213 				  unsigned long trans_exc_code)
214 {
215 	u16 instruction;
216 	int rc;
217 
218 	rc = __get_user(instruction, (u16 __user *) regs->psw.addr);
219 
220 	if (!rc && instruction == 0x0a77) {
221 		clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
222 		if (is_compat_task())
223 			sys32_sigreturn();
224 		else
225 			sys_sigreturn();
226 	} else if (!rc && instruction == 0x0aad) {
227 		clear_tsk_thread_flag(current, TIF_SINGLE_STEP);
228 		if (is_compat_task())
229 			sys32_rt_sigreturn();
230 		else
231 			sys_rt_sigreturn();
232 	} else
233 		do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code);
234 	return 0;
235 }
236 #endif /* CONFIG_S390_EXEC_PROTECT */
237 
238 static noinline void do_fault_error(struct pt_regs *regs, long int_code,
239 				    unsigned long trans_exc_code, int fault)
240 {
241 	int si_code;
242 
243 	switch (fault) {
244 	case VM_FAULT_BADACCESS:
245 #ifdef CONFIG_S390_EXEC_PROTECT
246 		if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
247 		    (trans_exc_code & 3) == 0) {
248 			signal_return(regs, int_code, trans_exc_code);
249 			break;
250 		}
251 #endif /* CONFIG_S390_EXEC_PROTECT */
252 	case VM_FAULT_BADMAP:
253 		/* Bad memory access. Check if it is kernel or user space. */
254 		if (regs->psw.mask & PSW_MASK_PSTATE) {
255 			/* User mode accesses just cause a SIGSEGV */
256 			si_code = (fault == VM_FAULT_BADMAP) ?
257 				SEGV_MAPERR : SEGV_ACCERR;
258 			do_sigsegv(regs, int_code, si_code, trans_exc_code);
259 			return;
260 		}
261 	case VM_FAULT_BADCONTEXT:
262 		do_no_context(regs, int_code, trans_exc_code);
263 		break;
264 	default: /* fault & VM_FAULT_ERROR */
265 		if (fault & VM_FAULT_OOM)
266 			pagefault_out_of_memory();
267 		else if (fault & VM_FAULT_SIGBUS) {
268 			do_sigbus(regs, int_code, trans_exc_code);
269 			/* Kernel mode? Handle exceptions or die */
270 			if (!(regs->psw.mask & PSW_MASK_PSTATE))
271 				do_no_context(regs, int_code, trans_exc_code);
272 		} else
273 			BUG();
274 		break;
275 	}
276 }
277 
278 /*
279  * This routine handles page faults.  It determines the address,
280  * and the problem, and then passes it off to one of the appropriate
281  * routines.
282  *
283  * interruption code (int_code):
284  *   04       Protection           ->  Write-Protection  (suprression)
285  *   10       Segment translation  ->  Not present       (nullification)
286  *   11       Page translation     ->  Not present       (nullification)
287  *   3b       Region third trans.  ->  Not present       (nullification)
288  */
289 static inline int do_exception(struct pt_regs *regs, int access,
290 			       unsigned long trans_exc_code)
291 {
292 	struct task_struct *tsk;
293 	struct mm_struct *mm;
294 	struct vm_area_struct *vma;
295 	unsigned long address;
296 	int fault;
297 
298 	if (notify_page_fault(regs))
299 		return 0;
300 
301 	tsk = current;
302 	mm = tsk->mm;
303 
304 	/*
305 	 * Verify that the fault happened in user space, that
306 	 * we are not in an interrupt and that there is a
307 	 * user context.
308 	 */
309 	fault = VM_FAULT_BADCONTEXT;
310 	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
311 		goto out;
312 
313 	address = trans_exc_code & __FAIL_ADDR_MASK;
314 	/*
315 	 * When we get here, the fault happened in the current
316 	 * task's user address space, so we can switch on the
317 	 * interrupts again and then search the VMAs
318 	 */
319 	local_irq_enable();
320 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address);
321 	down_read(&mm->mmap_sem);
322 
323 	fault = VM_FAULT_BADMAP;
324 	vma = find_vma(mm, address);
325 	if (!vma)
326 		goto out_up;
327 
328 	if (unlikely(vma->vm_start > address)) {
329 		if (!(vma->vm_flags & VM_GROWSDOWN))
330 			goto out_up;
331 		if (expand_stack(vma, address))
332 			goto out_up;
333 	}
334 
335 	/*
336 	 * Ok, we have a good vm_area for this memory access, so
337 	 * we can handle it..
338 	 */
339 	fault = VM_FAULT_BADACCESS;
340 	if (unlikely(!(vma->vm_flags & access)))
341 		goto out_up;
342 
343 	if (is_vm_hugetlb_page(vma))
344 		address &= HPAGE_MASK;
345 	/*
346 	 * If for any reason at all we couldn't handle the fault,
347 	 * make sure we exit gracefully rather than endlessly redo
348 	 * the fault.
349 	 */
350 	fault = handle_mm_fault(mm, vma, address,
351 				(access == VM_WRITE) ? FAULT_FLAG_WRITE : 0);
352 	if (unlikely(fault & VM_FAULT_ERROR))
353 		goto out_up;
354 
355 	if (fault & VM_FAULT_MAJOR) {
356 		tsk->maj_flt++;
357 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0,
358 				     regs, address);
359 	} else {
360 		tsk->min_flt++;
361 		perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0,
362 				     regs, address);
363 	}
364 	/*
365 	 * The instruction that caused the program check will
366 	 * be repeated. Don't signal single step via SIGTRAP.
367 	 */
368 	clear_tsk_thread_flag(tsk, TIF_SINGLE_STEP);
369 	fault = 0;
370 out_up:
371 	up_read(&mm->mmap_sem);
372 out:
373 	return fault;
374 }
375 
376 void __kprobes do_protection_exception(struct pt_regs *regs, long int_code)
377 {
378 	unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
379 	int fault;
380 
381 	/* Protection exception is supressing, decrement psw address. */
382 	regs->psw.addr -= (int_code >> 16);
383 	/*
384 	 * Check for low-address protection.  This needs to be treated
385 	 * as a special case because the translation exception code
386 	 * field is not guaranteed to contain valid data in this case.
387 	 */
388 	if (unlikely(!(trans_exc_code & 4))) {
389 		do_low_address(regs, int_code, trans_exc_code);
390 		return;
391 	}
392 	fault = do_exception(regs, VM_WRITE, trans_exc_code);
393 	if (unlikely(fault))
394 		do_fault_error(regs, 4, trans_exc_code, fault);
395 }
396 
397 void __kprobes do_dat_exception(struct pt_regs *regs, long int_code)
398 {
399 	unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
400 	int access, fault;
401 
402 	access = VM_READ | VM_EXEC | VM_WRITE;
403 #ifdef CONFIG_S390_EXEC_PROTECT
404 	if ((regs->psw.mask & PSW_MASK_ASC) == PSW_ASC_SECONDARY &&
405 	    (trans_exc_code & 3) == 0)
406 		access = VM_EXEC;
407 #endif
408 	fault = do_exception(regs, access, trans_exc_code);
409 	if (unlikely(fault))
410 		do_fault_error(regs, int_code & 255, trans_exc_code, fault);
411 }
412 
413 #ifdef CONFIG_64BIT
414 void __kprobes do_asce_exception(struct pt_regs *regs, long int_code)
415 {
416 	unsigned long trans_exc_code = S390_lowcore.trans_exc_code;
417 	struct mm_struct *mm = current->mm;
418 	struct vm_area_struct *vma;
419 
420 	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
421 		goto no_context;
422 
423 	local_irq_enable();
424 
425 	down_read(&mm->mmap_sem);
426 	vma = find_vma(mm, trans_exc_code & __FAIL_ADDR_MASK);
427 	up_read(&mm->mmap_sem);
428 
429 	if (vma) {
430 		update_mm(mm, current);
431 		return;
432 	}
433 
434 	/* User mode accesses just cause a SIGSEGV */
435 	if (regs->psw.mask & PSW_MASK_PSTATE) {
436 		do_sigsegv(regs, int_code, SEGV_MAPERR, trans_exc_code);
437 		return;
438 	}
439 
440 no_context:
441 	do_no_context(regs, int_code, trans_exc_code);
442 }
443 #endif
444 
445 int __handle_fault(unsigned long uaddr, unsigned long int_code, int write_user)
446 {
447 	struct pt_regs regs;
448 	int access, fault;
449 
450 	regs.psw.mask = psw_kernel_bits;
451 	if (!irqs_disabled())
452 		regs.psw.mask |= PSW_MASK_IO | PSW_MASK_EXT;
453 	regs.psw.addr = (unsigned long) __builtin_return_address(0);
454 	regs.psw.addr |= PSW_ADDR_AMODE;
455 	uaddr &= PAGE_MASK;
456 	access = write_user ? VM_WRITE : VM_READ;
457 	fault = do_exception(&regs, access, uaddr | 2);
458 	if (unlikely(fault)) {
459 		if (fault & VM_FAULT_OOM) {
460 			pagefault_out_of_memory();
461 			fault = 0;
462 		} else if (fault & VM_FAULT_SIGBUS)
463 			do_sigbus(&regs, int_code, uaddr);
464 	}
465 	return fault ? -EFAULT : 0;
466 }
467 
468 #ifdef CONFIG_PFAULT
469 /*
470  * 'pfault' pseudo page faults routines.
471  */
472 static ext_int_info_t ext_int_pfault;
473 static int pfault_disable = 0;
474 
475 static int __init nopfault(char *str)
476 {
477 	pfault_disable = 1;
478 	return 1;
479 }
480 
481 __setup("nopfault", nopfault);
482 
483 typedef struct {
484 	__u16 refdiagc;
485 	__u16 reffcode;
486 	__u16 refdwlen;
487 	__u16 refversn;
488 	__u64 refgaddr;
489 	__u64 refselmk;
490 	__u64 refcmpmk;
491 	__u64 reserved;
492 } __attribute__ ((packed, aligned(8))) pfault_refbk_t;
493 
494 int pfault_init(void)
495 {
496 	pfault_refbk_t refbk =
497 		{ 0x258, 0, 5, 2, __LC_CURRENT, 1ULL << 48, 1ULL << 48,
498 		  __PF_RES_FIELD };
499         int rc;
500 
501 	if (!MACHINE_IS_VM || pfault_disable)
502 		return -1;
503 	asm volatile(
504 		"	diag	%1,%0,0x258\n"
505 		"0:	j	2f\n"
506 		"1:	la	%0,8\n"
507 		"2:\n"
508 		EX_TABLE(0b,1b)
509 		: "=d" (rc) : "a" (&refbk), "m" (refbk) : "cc");
510         __ctl_set_bit(0, 9);
511         return rc;
512 }
513 
514 void pfault_fini(void)
515 {
516 	pfault_refbk_t refbk =
517 	{ 0x258, 1, 5, 2, 0ULL, 0ULL, 0ULL, 0ULL };
518 
519 	if (!MACHINE_IS_VM || pfault_disable)
520 		return;
521 	__ctl_clear_bit(0,9);
522 	asm volatile(
523 		"	diag	%0,0,0x258\n"
524 		"0:\n"
525 		EX_TABLE(0b,0b)
526 		: : "a" (&refbk), "m" (refbk) : "cc");
527 }
528 
529 static void pfault_interrupt(__u16 int_code)
530 {
531 	struct task_struct *tsk;
532 	__u16 subcode;
533 
534 	/*
535 	 * Get the external interruption subcode & pfault
536 	 * initial/completion signal bit. VM stores this
537 	 * in the 'cpu address' field associated with the
538          * external interrupt.
539 	 */
540 	subcode = S390_lowcore.cpu_addr;
541 	if ((subcode & 0xff00) != __SUBCODE_MASK)
542 		return;
543 
544 	/*
545 	 * Get the token (= address of the task structure of the affected task).
546 	 */
547 	tsk = *(struct task_struct **) __LC_PFAULT_INTPARM;
548 
549 	if (subcode & 0x0080) {
550 		/* signal bit is set -> a page has been swapped in by VM */
551 		if (xchg(&tsk->thread.pfault_wait, -1) != 0) {
552 			/* Initial interrupt was faster than the completion
553 			 * interrupt. pfault_wait is valid. Set pfault_wait
554 			 * back to zero and wake up the process. This can
555 			 * safely be done because the task is still sleeping
556 			 * and can't produce new pfaults. */
557 			tsk->thread.pfault_wait = 0;
558 			wake_up_process(tsk);
559 			put_task_struct(tsk);
560 		}
561 	} else {
562 		/* signal bit not set -> a real page is missing. */
563 		get_task_struct(tsk);
564 		set_task_state(tsk, TASK_UNINTERRUPTIBLE);
565 		if (xchg(&tsk->thread.pfault_wait, 1) != 0) {
566 			/* Completion interrupt was faster than the initial
567 			 * interrupt (swapped in a -1 for pfault_wait). Set
568 			 * pfault_wait back to zero and exit. This can be
569 			 * done safely because tsk is running in kernel
570 			 * mode and can't produce new pfaults. */
571 			tsk->thread.pfault_wait = 0;
572 			set_task_state(tsk, TASK_RUNNING);
573 			put_task_struct(tsk);
574 		} else
575 			set_tsk_need_resched(tsk);
576 	}
577 }
578 
579 void __init pfault_irq_init(void)
580 {
581 	if (!MACHINE_IS_VM)
582 		return;
583 
584 	/*
585 	 * Try to get pfault pseudo page faults going.
586 	 */
587 	if (register_early_external_interrupt(0x2603, pfault_interrupt,
588 					      &ext_int_pfault) != 0)
589 		panic("Couldn't request external interrupt 0x2603");
590 
591 	if (pfault_init() == 0)
592 		return;
593 
594 	/* Tough luck, no pfault. */
595 	pfault_disable = 1;
596 	unregister_early_external_interrupt(0x2603, pfault_interrupt,
597 					    &ext_int_pfault);
598 }
599 #endif
600